Visibility calculations for 3D computer graphics

Description

FIELD OF INVENTION
The present invention relates to computer graphics and, in particular, to the efficient determination of visible and/or invisible surfaces to thereby preferably permit improved hidden surface removal, generally in 3D systems.
BACKGROUND ART
Visible surface detection is one of the most basic operations in 3D graphics. It is applied to generate images of surfaces directly visible to a viewer. Recently, it has also been adopted in the radiosity calculations to compute the energy interactions between surfaces.
The standard strategy of visible surface detection is to divide surfaces into patch elements and compare the spatial relationship between these elements. Using this strategy, the visibility of surfaces cannot be determined until they have been analysed in detail. Although many techniques have been developed to address this issue, none are ideal as they either still require elaborate analysis of surfaces, or they impose various restrictions to the scene.
The limitations of the current techniques can seriously affect the speed of visible surface computation. If the scene is complicated, many surfaces may be invisible. However, the image generation is often slowed down by the need to analyse each surface element in detail.
The same limitation has also seriously affected the efficiency of the radiosity computations. Currently, these computations are very slow due to the need to elaborately compute the visibility between every pair of surface elements. Again, these computations may be substantially reduced if surface elements obviously visible or invisible to each other can be more easily computed.
The early visible surface techniques mainly applied various sorting schemes to find the occluding surface primitives. However, with the advancement in hardware technology, it is now common practice to reduce the need for sorting and comparisons by using a large amount of fast memory. This memory may be used to store the object data, such as a BSP-tree. It may also be used to store the depth and surface projection data, as with a z-buffer.
The z-buffer method is simple and has a very low growth rate. However, it still requires depth evaluations and comparisons at each pixel by all the patches projecting onto it, because their visibility is unknown.
The BSP-tree method has the advantage that if the scene is static, an orderly traversal of the BSP-tree is generally sufficient to establish the depth order. However, it still requires the scan-conversion of each path. In addition, the technique needs to re-compute the BSP-tree whenever objects in the scene move.
There have been two main strategies of avoiding detailed depth analysis of totally invisible entities. One strategy, applies the property that visibility changes can only occur at the contour edges of surfaces. Visibility computations of internal edges or patches can be reduced by first comparing them with these edges.
An alternative strategy is to use the invisible coherence of the scene. These techniques apply the property that an edge is likely to remain invisible in a scan line if it is invisible in the last scan line. Such an edge may therefore be treated specially to avoid unnecessary depth comparisons.
Although the above strategies can reduce some visibility computations, they have limitations. The contour-oriented techniques can operate only in environments which consist exclusively of convex or non-penetrating surfaces. Both the contour-oriented and the scan line approaches are also limited in their ability to reduce visibility computations. In the former, an edge still needs to be tested against the contour edges even if it is invisible. In the latter, all the invisible edge segments at each scan line have to be sorted. These segments also require depth comparisons with the pixels they are on. Finally, all these techniques require some sorting or searching.
SUMMARY OF THE INVENTION
It is an object of the present invention to substantially overcome, or ameliorate, the problems associated with the prior art, through provision of an improved method for performing visibility calculations.
In accordance with one aspect of the present invention there is disclosed a method of reducing the complexity of visibility calculations required for the production of multi-dimensional computer generated images or the reduction of multi-dimensional data to multi-dimensional data having at least oneless dimension, said method comprising the steps of:
(1) prior to an occlusion or invisibility relationship computation (known per se) being carried out on a plurality of surfaces, selected viewpoints to be calculated are divided into groups;
(2) for selected ones of said surfaces, determining for each said group whether each said selected surface is
(a) an always unoccluded surface, an always hidden surface, or a remaining surface; or
(b) an always unoccluded surface, or a remaining surface; or
(c) an always hidden surface, or a remaining surface; wherein said remaining surface is a surface which is unable to be determined with certainty as to whether it is either unoccluded or hidden;
(3) exempting from said occlusion or invisibility relationship computation those surfaces which are either always unoccluded or always hidden;
(4) maintaining a record of said remaining surfaces; and
(5) carrying out occlusion or invisibility relationship computations on said remaining surfaces.
In accordance with another aspect of the present invention there is disclosed a method of reducing the visibility related computations in an environment consisting of three-dimensional abstract or physical surfaces, said method comprising the steps of:
(1) prior to a visibility computation (known per se) being carried out, some or all viewpoints or possible occurrences of viewpoints are classified into groups of viewpoints;
(2) defining one or more projection surfaces (known per se) for the purpose of generating simultaneous projections of surfaces or surface elements with respect to a group of viewpoints;
(3) for selected surfaces, and for each group of viewpoints, determining whether those surfaces or their surface elements are always invisible to said group by computing and comparing their simultaneous projections on said projection surfaces;
(4) ignoring or treating specially some or all surfaces or surface elements which are always invisible to said group of viewpoints during the actual visibility or visibility related computations for some or all viewpoints in said group.
In accordance with another aspect of the present invention there is disclosed a method of reducing the visibility, radiosity or visibility related computations in an environment consisting of three-dimensional abstract or physical surfaces, said method comprising the steps of:
(1) prior to a visibility computation (known per se) being carried out, some or all viewpoints or possible occurrences of viewpoints are classified into groups of viewpoints;
(2) defining one or more projection surfaces (known per se) for the purpose of the simultaneous projections of surfaces or surface elements with respect to a group of viewpoints;
(3) dividing each of said projection surfaces into regular or irregular grids;
(4) defining a data structure which organizes computer storage for storing of projections on said grids;
(5) for each of the selected surfaces and for each group of viewpoints, simultaneously projecting said surfaces or their surface elements onto the projection surfaces, computing grid cells which are always under the projections, storing the furthest possible distances between viewpoints of said group and said surfaces to provide surface distances corresponding to those cells in said computer storage;
(6) for each patch element of said surfaces, determining those said grid cells that said patch element might project onto, comparing the depths stored in the cells with the furthest possible depth between the patch element and said group of viewpoints to determine whether the patch element is always occluded by other surfaces;
(7) ignoring or treating specially some or all surfaces or surface elements which are always visible to said group of viewpoints during the actual visibility or visibility related computations for some or all viewpoints in said group.

BRIEF DESCRIPTION OF THE DRAWINGS
A preferred and a number of other embodiments of the present invention will now be described with reference to the drawings in which:
FIGS. 1A and 1B show two viewpoint hierarchies;
FIG. 2 illustrates the projections at an arbitrary viewpoint;
FIG. 3 shows the relationship between projection boxes and a fuzzy projection box;
FIG. 4 shows the creation of fuzzy and non-fuzzy regions;
FIGS. 5A, 5B and 5C show a front-facing surface and its corresponding fuzzy and non-fuzzy regions;
FIG. 5D shows a point that is not totally hidden to a viewpoint bounding box;
FIG. 5E shows a point that is totally hidden to a viewpoint bounding box;
FIGS. 6A to 6D show the detection of totally invisible patches:
FIGS. 7A and 7B show the detection of totally visible/non-hiding patches:
FIGS. 8A to 8D show the scan-conversion of the non-fuzzy regions;
FIGS. 9A to 9D show the scan-conversion of the fuzzy regions; and
FIG. 10 illustrates the mapping between a cache and the fuzzy array.
FIGS. 11A and 11B show views of a front-facing surface region;
FIGS. 12A, 12B and 12C show cut-plane views of the projection of surface regions;
FIGS. 13A, 13B and 13C show respectively one, two and four room models used in testing the preferred embodiment;
FIG. 14 illustrates the direct computation of the fuzzy region of an edge;
FIG. 15 shows a hemi-cube pyramid structure;

BEST AND OTHER MODES FOR CARRYING OUT THE INVENTION
The preferred embodiments relate to various methods for calculating three dimensional computer graphic images and they are generally termed herein as the "fuzzy projection methods".
A Simple Fuzzy Projection Method
In this embodiment a method is provided to map the projections of entities at one or more viewpoints to a set of planes. This embodiment also provides a means to compute regions which contain all these projections and regions which are within each and every projection. The spatial position, time and the optical properties described in this embodiment can be replaced by other physical or abstract variables if the replacement does not affect the mathematical relationship of entities in the graphic image.
Prior to detailed discussion of this and the other embodiments, it is useful to introduce various terms and define variables used in 3D graphics.
Firstly, with reference to FIGS. 1A, 1B and 2, a viewer, as shown by the eyeballs in the Figs., is an abstract and dimensionless observer where the visibility of its surrounding objects is to be obtained. A point is visible to a viewer if a ray fired from the viewer to it is not blocked by any opaque object. There can be more than one viewer at any time and each viewer may move or rotate over the time. A viewpoint VP is the spatial position and orientation of a viewer at an instance of time. A view vector VV is a vector from the viewpoint VP to a point PO being observed (see FIG. 2).
Several viewpoints VP can be grouped together through certain properties they share. Related groups of viewpoints may in turn form a higher level group. This merging may be repeated to create a viewpoint hierarchy. A current group is the group of viewpoints VP that are currently being considered. In FIG. 1A, viewpoints VP.sub.1, VP.sub.2, VP.sub.3 and VP.sub.4 each reside on a path PA.sub.1 and represent viewpoints at times 1, 2, 3 and 4, and can be considered as one group. Similarly, for the viewpoints VP.sub.5 -VP.sub.7.
Each group of viewpoints is associated with a coordinate system called the group coordinate system CS. A viewpoint bounding box BB of a group of viewpoints VP is the smallest right quadrangular prism enclosing these viewpoints and whose edges are parallel to the axes of the group coordinate system. If the positions or occurrences of a group of viewpoints are not precisely defined, the associated viewpoint bounding box should be the said prism which encloses the space likely to be occupied by these viewpoints. A group of viewpoints may be lying on a plane or a point. In such cases, the viewpoint bounding box may degenerate into a plane or a point. A point PO is totally visible from the bounding box BB if it is always visible from every possible viewpoint VP in the box BB. Conversely, a point is totally invisible from the bounding box if it is hidden from the view of every possible viewpoint VP in the box BB. In FIG. 1A, viewpoints VP.sub.1 -VP.sub.4 reside in a bounding box BB.sub.1, and viewpoints VP.sub.5 -VP.sub.7 reside in a bounding box BB.sub.2. The boxes BB.sub.1 and BB.sub.2 can be considered as first level bounding boxes. If the boxes BB.sub.1 and BB.sub.2 are combined, a second level bounding box BB.sub.3 is formed. Rather than the bounding box being a square or rectangular prism, it can also be an ellipsoid, sphere or arbitrary volume.
A viewpoint group may be degenerated and contain only a single viewpoint. In this manner, all the techniques applicable to a normal viewpoint group are also applicable to a single viewpoint. In this case, the viewpoint bounding box BB degenerates into a point which coincides with the viewpoint VP.
In FIG. 1B, multiple viewers at different times are illustrated. A path PA.sub.2 of one viewer of a group is shown extending between locations at two points in time. VP.sub.8 represents the viewpoints at time "1" and VP.sub.9 the viewpoints at time "2", with BB.sub.4 the bounding box at time "1" and BB.sub.5 the bounding box of the union of the viewpoints VP.sub.8 and VP.sub.9.
Referring now to FIG. 2, shown is a projection box PB, a right quadrangular prism with its center on a viewpoint VP and its edges parallel to the principal axes of the current group coordinate system CS. The six facets of the box PB are called the projection faces PF. Each projection face PF is on a plane called a projection plane PP. The projection PR of a point PO.sub.1 in the direction perpendicular to a projection plane PP is the intersection between that plane PP and the view vector VV from the viewpoint VP to the point PO.sub.1. Conversely, since a straight line emanating from the viewpoint VP can only intersect the projection point PR on a projection plane PP, a location on that plane also represents a unique viewing direction from the viewpoint.
In this manner:
VV=PO.sub.1 -VP
The area around the visible space point PO.sub.1 is a visible patch PT and the view vector VV passes through this to an invisible space point IP, occluded by the patch PT.
The projection calculations on each plane are identical to the finding of the image region of an object in the direction perpendicular to that plane. The projection of an object to a projection plane PP therefore represents its image region in the direction parallel to the normal N of that plane.
A viewpoint VP may not be omni-directional but has a limited field of view. Hence certain projection faces may be partially or totally unobservable from the viewpoint. The hidden area of a projection face PF can be obtained by intersecting it with the viewing horizon of the viewpoint VP.
Locations on the projection plane PP can be expressed by a coordinate system CS called the projection coordinates. The origin of this coordinate system is at the center of the projection face PF. Its x, y axes are parallel to the edges and its z axis is parallel to the normal of the face.
Because they are all parallel, the projection coordinate systems of the current group of viewpoints VP have the same scale factor so that points having the same viewing direction from their viewpoints always have the same projection coordinates.
If points on two surfaces project on the same location, PR for example, on a projection face PF an occlusion is taking place. The point on the surface further from the viewpoint VP is hidden by the point on the surface closer to the viewpoint VP unless the latter is transparent. This can be determined by the distances d between these points PR and the viewpoint VP.
Every edge of the projection boxes PB of a group is parallel to one of the principal axes (x, y or z) of the respective group coordinate system CS. Since the vectors parallel to the principal axes can have at most sin directions, these faces are classified by their normal directions into six sets.
This is shown in FIG. 3 where viewpoints VP.sub.1, VP.sub.2 and VP.sub.3, each with respective projection boxes PB.sub.1, PB.sub.2 and PB.sub.3, reside within a bounding box BB.sub.1.
The projections on projection boxes PB.sub.1 -PB.sub.3 can be correlated by another box whose edges are also parallel to the axes of the current group coordinate system CS. This box is called the fuzzy projection box FB. It is called "fuzzy" because it is the combined projection of a single surface (or object, entity, etc.) from all viewpoints, with the combined projection creating a fuzzy image of the surface. Each face of the box is called a fuzzy projection face FF. The face FF lies on a plane called the fuzzy projection plane FP. A projection coordinate system can be defined for a fuzzy projection plane FP and the fuzzy projection face PF on it. Similar to the projection coordinate systems CS of the projection planes, the origin of this coordinate system is at the centre of the fuzzy projection face FF and its axes are respectively parallel to the edges and the normal of the face FF.
By setting the scale factor of the projection coordinate system of each of the fuzzy projection planes to be the same as the scale factor of their associated projection planes PP, each set of projection planes/faces has a one-to-one mapping with the fuzzy projection plane/face that is facing the same direction. Points on a projection plane/face can therefore be linearly mapped to the associated fuzzy projection plane/face.
Points on a set of projection planes having the same projection coordinates, and hence representing the same viewing direction, map to a point on the associated fuzzy projection plane that has the same coordinates. Therefore, similar to the projection planes PP, a point on the fuzzy projection plane FP also represents a unique viewing direction. However, the point is not associated with a particular viewpoint.
Since the viewpoints VP can have a limited field of view, some areas of the projection planes may be unobservable. The hidden area of a fuzzy projection face FF is the area where the corresponding areas on all the associated projection faces are hidden. A fuzzy projection face FF is inactive if all its area is hidden.
To sample the projections on the Fuzzy projection box FB, each fuzzy projection plane FP is tessellated by two sets of parallel and evenly spaced grid lines forming a fuzzy array FA. By ensuring that there are always grid lines on the edges of the fuzzy projection faces FF, these faces are divided into identical rectangular or square cells. Each cell is called a fuzzy element FE. Each element FE, in additional to representing the viewing direction associated with its center, also represents a unique solid angle A of viewing directions, as seen in FIG. 2, represented by its perimeter.
Other representation schemes such as the quadtree subdivision can be applied. All the z-buffer oriented operations would be accordingly changed to operation under such schemes, however the regular subdivision of the faces is considered the most efficient embodiment of the present invention.
Surfaces in the viewing environment are approximated by meshes of patches PT. A surface can also be represented by a hierarchy of patches. Different levels of the hierarchy are meshes of patches at different details. Each patch PT is treated as flat and is approximated by a polygon. The projection of a surface SU to a viewpoint VP is the region combining all the projection regions of its patches. This region is called the projection region PE of the surface. Since the projection region PE of a patch PT represents its image region, the projection region PE of a surface SU also represents the image region of that surface.
The smallest right quadrangular prism just enclosing the patch and whose edges are parallel to the axes of the current group coordinate system is called the bounding volume of the patch. The shortest and longest depths of a patch are the shortest and longest depths between the current viewpoints and points on the patch. The depth corresponds to the z-magnitude in the coordinate direction perpendicular to the projection planes currently being considered they are usually difficult to find, they are approximated by the shortest and the longest depths between the viewpoint bounding box and the bounding volume of the patch, which can be determined by partitioning the space by the planes of the viewpoint bounding box and determining which partitions the bounding volume of the patch is in.
This is shown in FIG. 4 where a patch PT of a surface is viewed from points VP.sub.1, VP.sub.2 and VP.sub.3 in a bounding box BB.sub.1. Associated with each viewpoint VP.sub.1 -VP.sub.3 is a projection region PE.sub.1 -PE.sub.3 which is seen to intersect with the corresponding projection face PF.sub.1 -PF.sub.3 in a different manner for each viewpoint VP.sub.1 -VP.sub.3.
The projections of the patch PT on the same set of projection faces PF.sub.1 -PF.sub.3 can be mapped to the associated fuzzy projection face FF. The areas on the fuzzy projection box FB.sub.1 containing all these mappings are called the fuzzy (combined) regions FR of the patch because logically the region is fuzzy. They cannot tell whether the patch PT.sub.1 actually projects on the corresponding area on each projection box PB.sub.1 -PB.sub.3.
Similar to the case of patches, the fuzzy region FR of a surface SU is produced by combining the mappings of the surface projections for a number of patches. In this manner each projection region of the surface is the logical-OR of all the projections of the patches, and the fuzzy projection of each patch is the logical-OR of its projection regions. Accordingly, the fuzzy region FR of a surface SU is also the logical-OR of the fuzzy projections of its patches. This region is equivalent to the superimposition of all the images of the surface seen from the current viewpoints.
On the fuzzy projection box FB.sub.1, there can also be areas which are always mapped on by the projection regions of the surfaces from the current group of viewpoints. On these areas the logical-AND of the projection regions PE.sub.1 -PE.sub.3 of the surface SU is true. Because the outcomes of the logical-AND operations are always a subset of the outcomes of the logical-OR operations, these areas are always within the fuzzy region FR of the surface SU.
Logically the areas are non-fuzzy as the projection status of the corresponding areas on the current projection boxes is always true. Therefore they are called the non-fuzzy region NF. An alternative term is the umbra region.
The non-fuzzy region NF of a surface SU can be obtained by mapping the projections of the surface at all the current group of viewpoints on the fuzzy projection box FB and finding the area within every projection. However, this is costly if there are many viewpoints. To reduce the computations, a series of approximations which err on the side of caution are applied.
It will be apparent to those skilled in the art that there can exist plural non-fuzzy regions for each surface.
Referring to FIG. 5A, the first approximation assumes that the viewpoints can be at any position in the current viewpoint bounding box BB.sub.2 and each viewpoint has a field of view containing all the possible fields of view of the current group of viewpoints. The patches PT.sub.n facing all these viewpoints are determined. They are called the front-facing patches. A method for finding these patches is described in detail below in sub-section headed The Detection of Patches Front-Facing to a Viewpoint Bounding Box.
Interconnected front-facing patches PT.sub.n are grouped into surface regions called the front-facing sub-surfaces SU.sub.2. The edges at the borders of these regions are called the boundary edges BE.
Since the front-facing patches PT.sub.n, and their boundary edges BE, are facing all the possible viewpoints VP in the viewpoint bounding box BB, a front-facing sub-surface SU.sub.2 never curves back and is always facing these viewpoints. The projections of its boundary edges BE therefore always surround its projection region PR.
Similar to the patches as shown in FIG. 5B, the projection of each boundary edge BE can be mapped on the fuzzy projection box FB.sub.1. A fuzzy region FR.sub.1 of the edge BE.sub.1 is the region on the box FB.sub.1 which contains all the projection mappings of that edge at every possible viewpoint in the current viewpoint bounding box BB.sub.2.
The fuzzy regions of all boundary edges belonging to a front-facing sub-surface can be combined into an area called the fuzzy boundary region BF. If a point is not in the fuzzy boundary region BF, the boundary edges BE do not project on all the associated points on the projection boxes. Because any change of projection status must be at the projection of the boundary edges, the point must be within every mapping of the projection regions of the front-facing sub-surface SU, or it must be outside all the mappings.
Since the fuzzy region is the combined projection of the front-facing sub-surface, any area within it must be mapped-by at least one projection region of the sub-surface. Therefore, the areas inside it but outside the fuzzy boundary region BF always contain the mappings of the projection regions of the sub-surface. These areas by definition are the non-fuzzy region NF.sub.1 of the front-facing sub-surface.
As the viewpoints can have any orientation and be anywhere in the viewpoint bounding box BB, the fuzzy region FR of an edge is difficult to compute. However, the extent of this region on a fuzzy projection face can be more easily obtained. The computation of this extent is described below in sub-section headed The Computation of the Fuzzy Extent of Boundary Edges and Patches. The smallest rectangle totally enclosing this extent and with boundary on the grid lines of the fuzzy projection plane is called the fuzzy extent EF of the edge on that plane.
This is seen in FIG. 5C where the same fuzzy projection face of FIG. 5B is shown in which the fuzzy boundary region BF.sub.1 is replaced by the fuzzy extents EF, of the boundary edges BE.sub.1 . . . etc.
The region containing all the fuzzy extents EF of the boundary edges always encloses the fuzzy boundary region of the front-facing sub-surface. Therefore, the subtraction of it from the fuzzy region FR produces an area always within the non-fuzzy region. This area is used to approximate the non-fuzzy region NF.
If a surface contains large patches and hence large boundary edges, the use of the fuzzy extents to approximate its fuzzy boundary region may be inefficient. The fuzzy regions of its boundary edges can be directly evaluated using the method described below in sub-section headed Direct Computation of the Fuzzy Region of an Edge. Alternatively, each edge can be divided into sub-edges. The fuzzy region of that edge can then be replaced by the fuzzy extents of its sub-edges.
An Invisibility Fuzzy Projection Method for the Detection of Totally Invisible Patches.
In this embodiment a method is provided to compare the mappings of entities on the fuzzy projection faces FF associated with a group of viewpoints VP. Through this operation, entities which may be treated as invisible to all these viewpoints, can be detected.
Referring to FIG. 6A, a viewpoint boundary box BB.sub.3 is shown to observe three patches PA.sub.A, PT.sub.B and PT.sub.C through a mesh surface SU.sub.M. FIGS. 6B, 6C and 6D show respectively the fuzzy extents EF.sub.A, EF.sub.B and EF.sub.C for the patches PA.sub.A, PT.sub.B and PT.sub.C on respective fuzzy projection faces FF.sub.A-C. Also shown is the non-fuzzy region NF.sub.3 of the mesh surface SU.sub.M.
The determination of totally invisible surfaces is based on a property of the non-fuzzy regions. In FIG. 5D, point PO2 is hidden by surface SU1 from viewpoint VP11 in the current viewpoint bounding box BB5. If the point is visible from another viewpoint VP10 which is also in BB5, then SU1 has to curve back and reach behind PO2. However, in such a situation part of SU1 next to a point SP would become back-facing to viewpoint VP12 in BB5. SU1 would then contain several front-facing surfaces which do not contain the surface around PO2. The non-fuzzy region of SU1 then would not cover the projection mapping of PO2.
Because of the above property, if the projection mapping of a point from an arbitrary point in the current viewpoint bounding box is within the non-fuzzy region of a surface, and if that point is behind the surface from the viewpoint, that point would always be hidden by the surface from all viewpoints in the box. This is shown in FIG. 5E. Here, the invisibility detection is performed using a single viewpoint (either one of VP13 or VP14 or other in the box BB6).
The above observation can be used to detect totally invisible patches, each patch whose visibility is to be determined is selected. The fuzzy elements within the projection mapping of the patch from a viewpoint in the viewpoint bounding box are accessed. If the distance stored in each element is smaller than the depth of the patch from the viewpoint in the direction associated with that element, the patch is an always invisible patch because of the property of non-fuzzy regions just mentioned.
To reduce the above computations, the projection mapping of the patch can be substituted by the extent of that mapping, or the fuzzy extent of the patch. Alternatively, the depth of the patch from the viewpoint may be replaced by the longest possible depth between the patch bounding box and the viewpoint or the viewpoint bounding box. Such approximations always err on the side of caution and do not affect the correctness of hidden surface computations.
2D fuzzy arrays FA.sub.A, FA.sub.B, FA.sub.C, with as many storage elements as the fuzzy elements EF on each active fuzzy projection face FF.sub.A-C are allocated to store the mapping information on the face. All the fuzzy arrays FA.sub.A-C are collectively called a fuzzy buffer generally, a memory that is two-dimensionally addressable.
Each fuzzy element EF contains a depth field. The depth field stores the depth of the patch PT whose fuzzy extent EF surrounds the fuzzy element FE.
A cycle of fuzzy buffer computations is applied for each group of viewpoints. First, the depth fields are initialized to infinity to indicate that no surface has yet been projected on the projection box PB.
The fuzzy extents EF of the boundary edges BE and the fuzzy extents EF of the patches PT of a front-facing sub-surface of selected opaque surfaces are computed (as described below in sub-section headed The Computation of the Fuzzy Extent of Boundary Edges and Patches). Usually, large surfaces are selected. Based on these extents, the non-fuzzy region NF of the front-facing sub-surface is computed by a scan line technique described below in sub-section headed The Computation Non-Fuzzy Regions. Furthermore, the fuzzy region of an edge can be found by direct computation as described in Appendix 6.
Using the above, the longest depth of the fuzzy elements in each horizontal segment of every non-fuzzy region are obtained. The distance stored in each element of the array is compared with the depth stored in the corresponding element in the fuzzy buffer. If the latter is larger, it is replaced by the former.
After all the non-fuzzy regions of the opaque front-facing sub-surfaces have been updated to the fuzzy elements, each element in the fuzzy buffer contains the longest depth of the closest surface that can be seen from a chosen viewpoint in the current viewpoint bounding box in the direction associated with that element. The fuzzy buffer can then be used to compute patches totally invisible from the current viewpoint bounding box. These patches are deemed to be totally invisible patches.
To detect totally invisible patches, the fuzzy elements within the fuzzy extent, the projection mapping, or the extent of the mappings of each patch are accessed. If the depth stored in each element is smaller than the shortest depth of the patch, the patch is always hidden in the direction represented by that element. Because of the property of non-fuzzy regions, or because of the property that the fuzzy extent covers all the possible viewing directions of the patch, the patch is totally invisible if it is found to be hidden at every fuzzy element within that extent.
As seen in FIG. 6B, the fuzzy extent EF.sub.A of patch PT.sub.A falls entirely within the non-fuzzy region NF.sub.3. Accordingly, the patch PT.sub.A is hidden from the viewpoints in the bounding box BB.sub.3.
In FIG. 6C, the fuzzy extent EF.sub.B of patch PT.sub.VB falls both inside and outside the non-fuzzy region NF.sub.3. Accordingly, those four elements outside may be either visible or hidden, and for the six elements inside, for all view directions through those elements, the patch PT.sub.B is hidden.
In FIG. 6D, the fuzzy extent EF.sub.C of patch PT.sub.C passes the depth test for every fuzzy buffer element it falls on. Patch PT.sub.C is always invisible from the bounding box BB.sub.3.
Since there is always substantial overlappings between the fuzzy extents of adjacent patches, a high-speed cache can be used to speed up the reading and writing of the fuzzy elements within the fuzzy extent of each patch. The cache can also be used in the visibility fuzzy projection technique discussed in the next section. Appendix 5 describes in more detail the use of a cache memory.
A Visibility Fuzzy Projection Method for the Detection of Totally Visible Patches.
In this embodiment a method is provided to compare the mappings of entities on the fuzzy projection faces associated with a group of viewpoints. Through this operation entities which may be treated as visible and not hiding other entitles to all these viewpoints can be detected.
For simplicity and speed, this method assumes that every surface does not hide itself in any viewing direction from the current viewpoint bounding box except those parts which do not face that direction. This implies that no two parts of the same surface that are facing a viewpoint in the bounding box would hide each other. All the planar and quadric surfaces have this property. Other surfaces can be subdivided into smaller surfaces which satisfy this criterion.
As seen in FIG. 7A, a viewpoint bounding box BB.sub.4 observes three patches PT.sub.D, PT.sub.E and PT.sub.F arranged about a mesh surface SU.sub.N.
Similar to the technique for detecting totally invisible patches, a fuzzy buffer consisting of several arrays is used to store the results of projections on the fuzzy elements.
Every element of the fuzzy buffer contains a field called the homogeneous indicator. This is a three-valued field which indicates whether there is any projection on the associated grid cell on the fuzzy projection box and whether one or more surfaces are being projected onto the cell. The indicator is zero if it is not enclosed by any fuzzy region. The indicator is one if it is inside the fuzzy region of one surface. The indicator is two if it is within the fuzzy regions of more than one surface. First, the homogeneous indicators are initialized to zero. This shows that no surface has yet been projected on the fuzzy buffer.
Using the scan-conversion technique described below in sub-section headed The Scan-Conversion of the Fuzzy Regions, the horizontal segments of the fuzzy region of each surface are obtained.
The homogeneous indicators under each segment are examined. Each indicator is incremented by one if it is zero or one.
After the fuzzy regions of all the surfaces have been updated into the fuzzy buffer, each patch is examined. It is ignored if it has been detected as totally invisible by the invisibility fuzzy projection technique.
If the homogeneous indicators within the fuzzy extent of the patch are all one, only one surface can be seen from the current viewpoint bounding box in all viewing directions within that fuzzy extent. The surface has to be the one the patch is on. The patch therefore does not hide other surfaces. Since the front-facing parts of the surface do not hide each other, the patch is always visible unless when it is back-facing. The patch is called a totally visible/non-hiding patch.
If some of the homogeneous indicators within the fuzzy extent of the patch are not one, the patch might be covering or be occluded by patches of other surfaces at certain viewpoints in the current viewpoint bounding box.
This is shown in FIG. 7B where the fuzzy-projection face FF.sub.4 is shown with each element having either the value 0, 1 or 2. As indicated for the patches PT.sub.D and PT.sub.F and their corresponding fuzzy elements FE.sub.D and FE.sub.F, these are not totally visible and not hiding other patches as some of the fuzzy elements within their respective fuzzy extents are mapped to more than one surface. Patch PT.sub.E is totally visible/non-hidden as all its fuzzy elements FE.sub.E within its extent are mapped only to one surface (i.e. SU.sub.N).
MODIFIED VISIBILITY FUZZY PROJECTION METHOD
The above technique scan-converts the fuzzy regions of each surface twice because each fuzzy element may be overlapped by the fuzzy regions of several patches of the same surface. This can be avoided by using a linked list called the projected patch list to store the overlapped patches for each fuzzy element. Other data structures such as an array can also be used in place of the linked list. The modified technique consists of the following steps:
1. As the original technique, each fuzzy element contains a homogeneous indicator and a surface ID field. It also contains a projected patch list. The homogeneous indicator and the surface ID fields are initialized to zero and null respectively to indicator that no surface has been projected on the fuzzy element. The projected patch list associated with the element is initialized to the null list.
2. Each patch contains a totally visible indicator. This indicator is initialized to ON. Each patch in the environment is accessed. The fuzzy region of each patch is computed and each element within it is accessed.
3. If the homogeneous indicator of each fuzzy element accessed is zero, the fuzzy element has not been projected onto by the fuzzy regions of any surface. The surface ID of the current patch is written to the surface ID field. The homogeneous indicator is set to one. If the totally visible indicator of the current patch is ON, its ID is added to the projected patch list of the current fuzzy element. Otherwise there is no such necessity because the projected patch list is used to reset the totally visible indicators of patches.
4. If the homogeneous indicator is one, the surface ID of the current patch is compared with the surface ID stored in the surface ID field of the current fuzzy element. If they are the same and the totally visible indicator of the patch is ON, the ID of that patch is added to the projected patch list associated with the current fuzzy element.
If the ID's of the two surfaces are different, the homogeneous indicator is set to two to indicate that there is more than one surface projecting onto the current fuzzy element. The current patch and all the patches with ID's stored in the projected patch list of the current element may not be totally visible. Therefore, their totally visible indicators are set to OFF. Records in the projected patch list can be removed after all their totally visible indicators have been updated.
5. If the homogeneous indicator is two, the current fuzzy element has already been found to be projected onto by more than one surface. The current patch therefore may not be totally visible. The totally visible indicator of its record is set to OFF.
6. After the processing of all patches, a patch is determined to be totally visible if its totally visible indicator has not been set to OFF. The pseudo-code of the above processing is shown below:
______________________________________initialize the totally invisible indicators of all patches to ON;initialize the homogeneous indicator of all fuzzy elements tozero;clear all projected patch lists of the fuzzy elements;for (each patch of the current surface)dofor (each fuzzy element under the fuzzy region of the patch)doif (the homogeneous indicator is zero)then write the surface ID of the current patch to the surface ID field; set the homogeneous indicator to one; if (the totally invisible indicator of the current patch is not OFF) add the patch ID to the projected patch list of the fuzzy element; endifelse if (the homogeneous indicator is one) then if (the surface ID of the current patch is the same as the surface ID stored in the fuzzy element) then if (the totally invisible indicator of the current patch is not OFF) add the patch ID to the projected patch list of the fuzzy element; endif else set homogeneous indicator to two; set the totally invisible indicator of the current patch and each patch indexed by the projected patch list of the fuzzy element to OFF; endif else set the totally invisible indicator of the current patch to OFF; endif done done______________________________________
Use of the Visibility and Invisibility Methods for the Radiosity Methods
By the application of the foregoing methods, a list of the totally invisible patches and the totally visible/non-hiding patches can be obtained for a viewpoint group. The visibility of these patches remains the same for all the subgroups and viewpoints under that group. Therefore these patches need not go through elaborate visibility computations in these subgroups and viewpoints.
For each viewpoint in the group, the totally invisible patches may be ignored. The totally visible/non-hiding patches need not be compared with other patches to determine their visibility and form-factors. Their form-factors can be computed directly from their orientations, positions and shapes using mathematical formulas describing the radiative transfer between surfaces.
If a patch is bright or close to the current group of viewpoints, the direct form-factor computations need to be carried out in higher accuracy. The patch can be subdivided into smaller patches and the form-factor computations are carried out for these sub-patches instead. Alternatively, accurate radiative transfer formulas such as the Nussel analog technique can be used. This technique finds the fractional area of the patch after it has been projected onto the surface and then the base of a hemisphere whose center is at the viewpoint.
If accurate form-factor computations are not required, the form-factor of a patch for the current group of viewpoints may be obtained by assuming that it is constant throughout that patch. The standard form-factor equation can be simplified to:
F=cos A*cos B*A/(.pi.r*r)
where A and B respectively are the angles between the view vectors and the normals of the observed patch, and the patch the viewpoint is on, and r is the distance between the viewpoint and the center of the observed patch.
The fuzzy projection methods require orderly top-down traversal of viewpoints according to their hierarchy. However, in techniques such as the progressive refinement methods, the receiving patches are usually accessed in decreasing order of their brightness. Therefore, if these techniques are used in conjunction with the fuzzy projection methods, the form-factor computations of some viewpoints in a group may not have been carried out since not all the viewpoints within the group may be processed in one go. Information about which are the totally visible/non-hiding patches and which are the totally invisible patches of the partially completed group has to be stored in memory. The outstanding viewpoints in the group can then use this information when their turn to be processed has arrived. This information can be removed when all the viewpoints in the group have been processed.
In techniques such as the progressive refinement methods (known per se), the form-factors between patches have to be repeatedly computed. Therefore, if memory permits, the information about which patches are totally visible/non-hiding and which patches are not totally invisible may not be removed and can be advantageously used.
In the first round of computations, all the fuzzy projection and normal hemicube computations for every receiving patch are carried out as required. During these computations, the emission patches at each group/viewpoint are classified into three groups: totally visible/non-hiding, totally invisible, and the remainder. This classification information is stored using data structures such as arrays or linked lists.
In subsequent rounds of computations, where the form factors between patches have to be re-evaluated, the fuzzy projection computations are not repeated. The totally visible/non-hiding and the totally invisible patches of each viewpoint can be simply retrieved from the stored information.
The memory for storing the classification of patches for each group of viewpoints can be reduced by several methods. First, since patches found to be totally visible/non-hiding or totally invisible from a group will maintain the same status for all its subgroups, they need not be repeatedly stored.
Secondly, because the union of the three lists of patches is the set of all the patches that need to be classified for each group of viewpoints, only two lists need to be stored for the group. To further reduce storage, the smallest pair of the three lists can be kept.
Also, because the records of patches are usually orderly organized and accessed, the patch IDs in the lists usually have some order. The differences between the IDs of successive patches in the lists are often much smaller than the magnitude of these IDs. Whenever this occurs, these differences instead of the patch IDs can be stored. A sign bit can be used to distinguish between the ID's and difference values. The list can be sorted by the ID for more efficient use of this arrangement.
Finally, the visibility status of patches often does not change much from one group of viewpoints to an adjacent group of viewpoints. Therefore, using the patch lists of a viewpoint group as a starting point, a series of nearby groups can store their patch lists incrementally. Starting from the first group, each adjacent group only needs to store the lists of patches that need to be deleted from, and added to the patch lists of the previous group to form the current lists. Again, a special field can be used to distinguish the use of this arrangement.
ENHANCING THE ACCURACY OF FORM-FACTOR COMPUTATIONS
The fuzzy projection techniques can also be used to achieve optimal accuracy of the form-factor computations. After the filtering off of totally invisible patches by the invisibility techniques the visibility technique detects totally visible and non-hiding patches. As mentioned above, different methods which yield different accuracy in the computations of the form-factors of these patches can be applied.
The remaining patches are not totally invisible. They are also not totally visible/non-hiding. The accuracy of their form-factor computations can be determined by the following steps:
a. First, several levels are defined such that each of them corresponds to a level of accuracy to be reached by the form-factor computations of patches. Each accuracy level determines the strategies of form-factor computations such as the resolution of the hemicube buffer, the level of details of patches, whether to use ray-tracing instead of the hemicube buffer, or the number of rays traced per sampling point.
b. All the patches not found to be totally invisible or totally visible are classified by their accuracy level according to their brightness, importance, distance from the current group of viewpoints, and other properties. Each patch record has a field which stores its accuracy level. An accuracy level field is also allocated for each fuzzy element in the fuzzy buffer. Each field is initialized to the lowest accuracy level.
c. Patches are accessed in the order such that those having higher accuracy levels are scan-converted first.
For each patch accessed, the same scan-conversion of their fuzzy extents applied in the visibility fuzzy projection computations is carried out. However, instead of the homogeneous indicators, the accuracy level fields in each fuzzy element access is examined. If the value in a field is larger than the accuracy level of the patch, the patch is likely to be hiding or be hidden by a patch at a higher accuracy level. The accuracy level of the patch is set to this value. If the value is smaller than the original accuracy level of the patch, the later is written to the former.
d. After the processing of all patches, the updated accuracy level of each patch shows the maximum level of accuracy to which its form-factor computations needs to be carried out for the current group of viewpoints. If this level is higher than the accuracy the patch can provide, it may be recursively subdivided and replaced by sub-patches whose accuracy matches the level.
If the actual surface of a patch is curved, the projections of its sub-patches may be outside its projection. Therefore, for the subdivision strategy to be effective, the fuzzy extent of each entity should contain the fuzzy extents of all its sub-entitles during all the fuzzy projections. It can be approximated by the fuzzy extent of the axis-aligned bounding box containing all these sub-entities.
e. The accuracy level field of a fuzzy element indicates the accuracy of the computations that needs to be carried out in the corresponding viewing directions from the current group of viewpoints or the corresponding regions on the hemicubes of these viewpoints. The form-factor of a patch can be computed in variable resolutions which match the different accuracy requirements of the fuzzy elements covered by that patch. This can be done in either the ray-tracing or the hemicube approach of form-factor computations.
f. In the ray-tracing approach, the form-factors of patches at a viewpoint are computed by tracing rays in all observable directions. For each of the current group of viewpoints, the number of rays to be fired within a solid angle associated with a fuzzy element depends on the accuracy level of that element. If the level is high, more than one ray should be traced. Conversely, if the accuracy levels of several adjacent fuzzy elements are low, only one ray may be traced for the directions associated with these elements.
g. In the hemicube approach, different regions on the hemicube should have different sampling resolutions which match the accuracy requirements of the fuzzy elements associated with these regions. This can be achieved by the use of a pyramidal representation called the hemicube pyramid. The scan-conversion of patches on the hemicube pyramid is described below in sub-section headed Scan-Conversion of Patches on the Hemicube Pyramid.
Use of the Visibility and Invisibility Methods for Hidden Surface Computations
In this embodiment a method is provided to apply the computed results of the visibility and invisibility fuzzy projection methods in the hidden surface computations.
In shadow computations, in applications where a series of images are to be taken, such as computer animation, flight simulation, or dynamic graphics, the hidden surface removal operations need to be repeated many times. The position of the viewer when an image of the surrounding is taken may be treated as a viewpoint. The viewpoints may also be combined into a hierarchy of viewpoint groups.
If the combined field of view of the current viewpoint group is narrow, the projection boxes and the fuzzy projection box can be flattened so that only a fuzzy projection plane and its associated projection planes are active for the current group of viewpoints.
By using the invisibility fuzzy projection methods patches found to be totally invisible to a group of viewpoints, do not need to be considered for the subgroups and the individual viewpoints in the group.
By using the visibility fuzzy projection method no depth computations and comparisions of totally visible patches during the scan-conversion are necessary.
Use of the Visibility and Invisibility Methods for the Ray Tracing Computations
In this embodiment a method is provided to apply the computed results of the visibility and invisibility fuzzy projection techniques in the ray tracing techniques.
In a ray tracing application, the objects and surfaces in the environment are often organized as a hierarchical geometric model. If the number of rays traced from a node of the geometric model justifies the use of the fuzzy buffer methods, a viewpoint group is created by treating every point on the surfaces of that node where a ray might emit as a viewer. The position of the viewer at any instance of time is a viewpoint. The viewpoints can be grouped into a hierarchy of viewpoint groups.
From the fuzzy buffer computations, patches which are totally visible/non-hiding to a group of viewpoints corresponding to each node and patches which are totally invisible to this group of viewpoints.
When a ray is traced from a node, all the totally invisible patches need not be considered. Also, the ray is tested first with the totally visible/non-hiding patches. If the ray hits one of those patches, no further testing with other patches need to be carried out as the patch being hit is the closest one on the path of the ray.
If both the radiosity method and the ray tracing techniques are used, the results of the fuzzy buffer computations can be used for both methods.
USE OF THE VISIBILITY AND INVISIBILITY FUZZY PROJECTION TECHNIQUES IN COMPUTER VISION
The visibility and invisibility fuzzy projection techniques can be used in computer vision. The occlusion analysis of objects is a vital operation in computer vision. Based on this analysis, hypotheses are made to construct a 3D model. This model is then matched with the vision data or existing models. It may be repeated matched and refined until it is acceptable. Usually there is a lot of uncertainty and vagueness in the vision data and the hypothesis. Such imprecision can be accommodated by the present techniques in the following ways:
1. During the model construction phases, a viewpoint bounding box can be defined which includes all the likely positions of the viewpoint.
2. During the model construction phase, a 3D model is generated based on the vision data. However, if the exact locations and shapes of entities such as edges and patches are uncertain, the entities may be approximated by the bounding boxes which contain all their likely locations.
3. Based on the hypothesized model, the invisibility and visibility techniques are carried out to compute totally visible surfaces, totally invisible surfaces and surfaces whose visibility cannot be determined.
4. The totally visible surfaces correspond to the vision data whose information is relatively explicated. Such data may be treated as successfully interpreted or more cursorily checked. Areas on the fuzzy projection box which are projected onto by surfaces whose visibility cannot be determined correspond to the vision data which are more obscured. The areas on the image plane corresponding to these areas are further analysed.
5. In the next round of analysis, the hypothesis becomes more refined. The viewpoint bounding box and bounding boxes of entities may be accordingly shrunked.
The preferred embodiment has been implemented on a general purpose computer adapted via programming in the C-language and tested using a set of models which differ in depth complexity. Three of the models are shown in FIGS. 13A, 13B and 13C. The visible surface computations have been computed for five nearby viewpoints. The times of hidden surface computations under different pixel and patch resolutions have been measured and are shown below.
EXAMPLE 1
Coarse patches, coarse screen pixels (400.times.400 pixels), resolution of fuzzy buffer: 400.times.400 pixels
______________________________________Normal Hidden Fuzzy HiddenSurface Removal Surface Removal CPU time CPU time Patchs for 5 for fuzzy CPU time invisible viewpoints Total computa- for 5 view- to allModel (sec) Patches ions(sec) points(sec) viewpoints______________________________________1 room 11.2 10807 1.2 11.1 1572 room 21.8 19137 2.1 17.3 39423 room 35.7 27467 2.9 23.9 90414 room 46.3 35797 3.6 25.6 16140______________________________________
EXAMPLE 2
Coarse patches, fine screen pixels (1000.times.1000 pixels), resolution of fuzzy buffer: 400.times.400 pixels
______________________________________Normal Hidden Fuzzy HiddenSurface Removal Surface Removal CPU time CPU time Patchs for 5 for fuzzy CPU time invisible viewpoints Total computa- for 5 view- to allModel (sec) Patches ions(sec) points(sec) viewpoints______________________________________1 room 20.7 10807 1.2 20.7 1572 room 40.8 19137 2.1 32.4 39423 room 56.4 27467 2.9 37.8 90414 room 72.9 35797 3.6 40.0 16140______________________________________
EXAMPLE 3
Fine patches, coarse screen pixels (400.times.400 pixels), resolution of fuzzy buffer: 400.times.400 pixels
______________________________________Normal Hidden Fuzzy HiddenSurface Removal Surface Removal CPU time CPU time Patchs for 5 for fuzzy CPU time invisible viewpoints Total computa- for 5 view- to allModel (sec) Patches ions(sec) points(sec) viewpoints______________________________________1 room 25.0 21263 2.0 24.1 16052 room 41.1 38293 2.5 25.0 153013 room 60.6 55323 3.4 29.1 287124 room 80.7 72353 4.6 33.8 42103______________________________________
EXAMPLE 4
Fine patches, fine screen pixels (1000.times.1000 pixels), resolution of fuzzy buffer: 400.times.400 pixels
______________________________________Normal Hidden Fuzzy HiddenSurface Removal Surface Removal CPU time CPU time Patchs for 5 for fuzzy CPU time invisible viewpoints Total computa- for 5 view- to allModel (sec) Patches ions(sec) points(sec) viewpoints______________________________________1 room 42.0 21263 2.0 40.5 16052 room 75.9 38293 2.5 45.4 153013 room 111.3 55323 3.4 53.8 287124 room 148.3 72353 4.6 61.9 42103______________________________________
The above results indicate that the overhead of the preferred embodiment is low. It also indicates that substantial computation savings can be achieved when the depth complexity of the models is high.
There are a number of advantages to be gained by applying the strategies in the disclosed embodiments. Firstly, they do not have the restrictions encountered by the earlier hidden surface algorithms. For example, they do not preclude the presence of intersecting surfaces or patches in the environment. In addition, surface patches can be curved and the scene need not be static. The method also has very little computational overhead. Because it operates on a z-buffer, it can be easily implemented in hardware.
The preferred embodiment can be further enhanced using heuristics. In applications such as the interactive walkthrough of the interior of a building, the surfaces most likely to form large crisp (non-Fuzzy, umbra) regions are those representing walls, floors and ceilings. Therefore, only the crisp regions of these surfaces need to be scan-converted to the fuzzy buffer.
In addition to reducing hidden surface removal computations, the preferred method can also be used in the radiosity method. Currently the radiosity method is very slow, mainly because it needs to compute the visibility of surfaces for a large number of surface points to determine the form-factors. However, by grouping these points and treating them as groups of viewpoints, the preferred embodiment can be used to determine surfaces totally visible to a group of these points. The fuzzy computations for each viewpoint bounding box are apportioned among all viewpoints in the box. Because there would be a viewpoint for every surface point, the density of viewpoints in each viewpoint bounding box could be very high. Therefore, the advantage of using the method in the radiosity method can be more than in the normal hidden surface removal.
The methods disclosed can also be used in virtual reality applications and other applications where abstract data is utilized. In the future such an application could need to display complex and dynamic environments in real time using the radiosity method. With the addition of a time dimension, the density of viewpoints in each viewpoint bounding box could be an order of magnitude higher than that of the radiosity method. Consequently, the methods disclosed could substantially speed up these applications. Other applications include the manipulation of physical data suc as the generation, processing and display of scientific data such as energy spectra data.
The foregoing describes only a number of embodiments of the present invention and modifications, obvious to those skilled in the art, can be made thereto without departing from the scope of the present invention.
The Detection of Patches Front-Facing to a Viewpoint Bounding Box
Referring to FIG. 2, whether a patch PT is facing a viewpoint VP can be determined by the angle A between the normal N of the patch PT and the view vector VV. The latter is the vector from the viewpoint VP to the center of the patch PT The cosine of the angle is: ##EQU1## where VP=(x, y, z) is the position of the viewpoint, N=(Nx, Ny, Nz) is the unit normal of the patch, and PO=(x', y', Z') is the position of the patch centre. k is a positive coefficient which turns the view vector (VV=PO-VP) into a unit vector.
If the patch PT is facing the viewpoint bounding box BB, the angle A must be obtuse. Hence cosine A must be negative for every possible viewpoint VP in the box. To check this, the maximum value of the cosine of A is found. If it is positive, then the patch would be back-facing to certain viewpoints in the bounding box.
Since PO and N are constant vectors, the maximization of cosine A requires each of Nx x, Ny y and Nz z to be minimized. Depending on the signs of the components of N, either the maxima or the minima of x, y and z are used. These coordinates correspond to the locations of the corners of the current viewpoint bounding box.
If a patch moves or rotates over time and the viewpoint bounding box contains viewpoints at different times, then the smallest cosine A at each instance of time may be computed and the smallest of them chosen.
Patches with bounding volumes inside or intersecting the viewpoint bounding box are always treated as not front-facing. These patches can be detected by an initial bounding volume check.
FIG. 11A shows a cut-plane view of a surface region SU.sub.R front-facing a set of viewpoints VP.sub.7, VP.sub.8, VP.sub.9 and FIG. 11B shows the same surface region front-facing to a viewpoint bounding box BB.sub.R. Since the front-facing-patches PT.sub.F are facing all possible viewpoints in the box BB.sub.R, the region SU.sub.R never curves back and is always facing these viewpoints. The view projections of the border of the region SU.sub.R therefore always enclose its projection. This property can be used in subsequent fuzzy projection computations.
FIGS. 12A-12C show cut-plane views of projections of surface regions front facing to a group of viewpoints and their respective bounding boxes.
In FIG. 12A if all patches PT.sub.A in a surface region are front facing, the projections BP.sub.A of the boundary edges of the region always surround the projection PE.sub.A of that region.
In FIG. 12B if some surface patches PT.sub.B in a surface region are not front facing, the region may curve back. The projections BP.sub.B of the boundary edges of the region may be inside the projection PE.sub.B of that region.
In FIG. 12C the projection BP.sub.C of the boundary of each front-facing region of a convoluting surface still encloses the projection PE.sub.C of the region.
The Computation of the Fuzzy Extent of Boundary Edges and Patches
As shown in FIG. 2, the projection coordinates PR (X,Y) of a point PO (x',y',z') to a viewpoint VP (x,y,z) on the projection plane PP facing the positive z direction is
X=(x'-x)*d/(z'-z) (1)
Y=(y'-y)*d/(z'-z) (2)
where d is the shortest distance from the viewpoint to the projection plane facing the z direction.
To maximize X and Y, the minimum values of x and y are used. If X and Y are to be minimizes the maxima of x and y are used. Since z is part of the divisors, whether to use its maximum or minimum value depends on the signs of (x'-x) and (y'-y) after choosing the extreme values of x and y. Since the edges of the viewpoint bounding box are parallel to the axes, viewpoints having these extreme x, y and z values are at the corners of the box.
According to equations (1) and (2), the points which have the extreme values in x', y' and z' would yield the maximum and minimum projections. However, these points might not be on the current projecting entity, which can be a boundary edge or a patch.
To overcome the problem, the bounding volume of the entity is used to approximate that entity. Its projections can be used as an estimation as they enclose the projections of that entity.
Sometimes an entity may be subdivided into sub-entities. Its position and shape may also be non-static over time or uncertain. Hence the bounding volume of an entity should be a volume that encloses all the space it is likely to occupied, and all the bounding volumes of its sub-entities that are likely to be considered.
Hence the bounding volume of an entity should enclose all its sub-entities that are likely to be considered in the computations.
Following the derivations above, the maxima of x' and y' and the minima of x and y are used to maximize the projections. Whether to maximize or minimize (z'-z) depends on the signs of (x'-x) and (y'-y). If these terms are positive, the term should be minimized. Hence the smallest z' and the largest z are used. Otherwise, the largest z' and the smallest z are instead chosen.
The finding of the minimum projections is the opposite of maximization. Hence the decisions to minimize or to maximize the individual terms are accordingly changed.
Having obtained the projections which give the maximum and minimum X and Y values, these projections are mapped to the fuzzy projection plane FP whose normal is toward the z direction.
A rectangle on the plane with these projections as corners and with edges parallel to the edges of the corresponding fuzzy projection face FF can be defined. They are either rounded up or down to create another region which contains the original rectangle and all the pixels under it. This is the fuzzy extent EF of the patch on the fuzzy projection plane FP.
By rotating the axes, the same equations (1) and (2) can be used for projection planes facing other directions if the terms in equations (1) and (2) are accordingly permuted. The fuzzy extents corresponding to these planes can then be similarly obtained.
If a patch is within the viewpoint bounding box, then for some viewpoints z' is smaller than or equal to z in equations (1) and (2). The fuzzy extents would then extend to infinity in the directions which can be determine by the signs of the denominators and numerators in these equations.
The Computation of the Non-Fuzzy Regions
To compute the non-fuzzy region, it is first necessary to obtain its relationship with the fuzzy boundary region and one of the projection regions.
Consider two viewpoints in the current viewpoint bounding box and a point on a fuzzy projection plane. Assume that the projection of the front-facing sub-surface at one viewpoint maps on that point and at the other viewpoint does not map on that point.
Because the change of projection status can only occur at the projections of boundary edges, on a curve joining the two points there must exist another viewpoint such that the projection of a boundary edge from it would map on the same point on the fuzzy projection plane. Since this curve can be totally inside the current viewpoint bounding box, that point must be within the fuzzy region of the boundary edges for that bounding box.
If a point on a fuzzy projection plane is within the mapping of a projection region of the front-facing sub-surface, and if it is outside the fuzzy region of the boundary edge, then the point is always within the projection regions of the sub-surface for all the viewpoints in the current viewpoint bounding box. The point is by definition inside the non-fuzzy region. Since the fuzzy boundary region is always within the area containing the fuzzy extents of boundary edges, the point is still inside the non-fuzzy region if the former is approximated by the latter.
Based on the above, the approximated non-fuzzy region can be obtained by finding the projection region of the front-facing sub-surface from a viewpoint in the current viewpoint bounding box projecting it to the fuzzy region, and subtracting all its area intersected with the fuzzy extents of the boundary edges. A modified scan line algorithm may be used for this task.
First, an arbitrary viewpoint in the current viewpoint bounding box is selected. Usually the center of the bounding box is chosen. This viewpoint is called the chosen viewpoint. The projection coordinates of the edges in the front-facing sub-surface from this viewpoint are found. Since the projection coordinate system and the fuzzy projection coordinate system have one-to-one mapping, the mapping of the former to the latter is trivial.
This is followed by the computations of the fuzzy extents of the patches described below in sub-section headed The Computation of the Fuzzy Extent of Boundary Edges and Patches.
An illustration of such an arrangement is shown in FIG. 8A where two patches A and B intersect a scan line SL and have respective fuzzy extents EF.sub.A and EF.sub.B of their boundary edges.
Two active edge lists are maintained during the processing of scan lines. The first list contains the active edges of the patches. The second list contains the pairs of active vertical edges of the fuzzy extents of boundary edges.
In FIG. 8A, the first active edge list is:
A1 A2 B1 B2
the second active edge list is:
E1 E2 E3 E4
During the processing of successive scan lines, entries in the list are added when they become active in the current scan line. They are deleted from the list when the last scan lines in which they are active have been processed.
A scan line array with as many elements as the number of horizontal elements in the current fuzzy array is maintained. Each element of the array contains a depth field and a boolean field called the overlap indicator. The depth field is used to store the depth of the patch whose fuzzy extent surrounds the fuzzy element. The overlap indicator indicates whether the element is under the non-fuzzy region. The depth fields are initialized to negative infinity. The overlap indicators are set to zero.
The pairs of edges in the first active edge list are then processed. The span of each pair of these edges represents the segment of a patch in the current scan line. The overlap indicators of the elements within the span are set to one. The value stored in the depth field of each of these elements is compared with the depth of the patch with respect to the chosen viewpoint. It may be approximated by the maximum depth from the viewpoint to the patch or the patch bounding box. If the former is smaller, it is replaced by the latter.
The computation of the segments of patches requires the finding of the intersections between the current scan line and the active edges. This computation can be avoided by using the leftmost or rightmost position of each edge depending on whether it is at the left or right side of the span. This inaccuracy is allowed because it is always offset by the fuzzy extents of the boundary edges.
The pairs of active edges in the second active edge list are then accessed. The span of each pair represents the segment of the fuzzy extent of a boundary edge in the current scan line. All the overlap indicators of the elements within the span are reset to zero.
After all the records in the active edge lists have been processed, the elements in the scan line array where the overlap indicators are one are within the segment of the non-fuzzy region in the current scan line. The depth fields contain the maximum possible depth of the region in the viewing directions represented by these elements.
The depth stored in each element of the array is compared with the depth field of the corresponding element in the fuzzy buffer. If the latter is larger, it is replaced by the former.
Returning to the Figs., FIG. 8B shows the scan line array after scan-conversion of the active edges of patch A. FIG. 8C shows the scan-conversion of the active edges of patch B. FIG. 8D shows the scan line array after the elements within the fuzzy extents EF.sub.A and EF.sub.B of the boundary edge have been reset.
The Scan-Conversion of the Fuzzy Regions
To obtain the fuzzy regions, a modified scan line algorithm to that of Appendix 3 is used.
An active edge list is maintained during the processing of scan lines. The list contains pairs of vertical edges of the fuzzy extents of patches active in the current scan line. During the processing of successive scan lines, entries in the list are added when they become active in the current scan line. An entry is deleted from the list when the last scan line it is active has been processed.
For the arrangement of FIG. 9A (similar to FIG. 8A, but for fuzzy regions), the active edge list is:
E1 E2 E3 E4.
A scan line array with as many elements as the number of horizontal elements in the current fuzzy array is maintained. Each element of the array is a boolean field called the overlap indicator which indicates whether the element is under the fuzzy region.
For each scan line, the entries in the active edge list are accessed. The span between each pair represents the segment of the fuzzy extent of a patch in the current scan line. All the overlap indicators of the elements within the span are set to one.
After the processing of the entries in the active edge list. The array elements whose overlap indicators are one now contain the segment of the fuzzy region in the current scan line.
FIG. 9B shows the state of the scan line array after the elements within the fuzzy extent EF.sub.A of patch A have been updated.
FIG. 9C shows the state of the scan line array after the elements within the fuzzy extent EF.sub.B of patch B have been updated.
Using a Cache to Speed up the Access and Writing of Fuzzy Elements Within the Fuzzy Extent of a Patch
The fuzzy buffer methods described herein require the reading and writing of the fuzzy elements within the fuzzy extent of each patch. Since there are usually substantial overlaps between the fuzzy extents of adjacent patches, the access and update of these elements can be reduced by storing their information in high speed cache.
The cache may be in many forms. One example, shown in FIG. 10 is to use a cache buffer CB and two cross-reference tables CR.sub.1 and CR.sub.2. The cache buffer CB is a 2D array, in which each element contains fields which are the mirror of the fields on a fuzzy buffer element that need to be read or written. The cross-reference tables CR.sub.1 and CR.sub.2 respectively contain as many elements as the number of rows and columns in the 2D array. Depending on the array it is in, each element in the table contains either a row (CR.sub.1) or a column (CR.sub.2) field which is used to cross-reference the rows and columns of the 2D array with the rows and columns of the current fuzzy array FA. For each element in the cache, the cross-reference tables CR.sub.1 and CR.sub.2 contain a reset switch to indicate whether it has been reset.
Initially, all the reset switches are initialized to one. The centre of the cache buffer CB is mapped to the center of the fuzzy extent of the first patch to be read. After the fuzzy elements within the fuzzy extent of the first patch are read, information in these elements are stored in the cache buffer CB elements according to the mapping. The mappings of rows and columns in the cache are updated into the cross reference table CR.sub.1 and CR.sub.2, respectively. The reset switch in the updated elements in these table and the cache buffer CB are set to zero.
Before reading the fuzzy elements within the fuzzy extent of the second patch, data in the cache buffer CB is first read. If an element has already been stored in the cache buffer CB, no access to the cache buffer CB to obtain its information is necessary.
To maximize the efficiency of the cache, any adjacent patch is accessed first. After the processing of a series of patches, there may be a patch whose mapping covers the boundary of the cache. The affected rows or columns are wrapped around. The rows at the opposite edges become the next rows or columns for the patch.
Before the writing of the patch data into the cache buffer CB, each element in the cross-reference table CR.sub.1 and CR.sub.2 within the rows of the columns of the current patch mapping are checked. If the reset switch is on or if the mapping value conforms to the current mapping, the whole row or column of cache elements have not been used and there is no problem in using them. Otherwise, the row or the column needs to be reclaimed. The original data in the whole row or column of cache elements are written back to the fuzzy buffer and then initialized. The affected element in the cross-reference table is updated with the new mapping. After that, the patch data can then be written into these newly-reclaimed cache elements.
The writing from the cache buffer CB to the fuzzy buffer can be performed in burst mode. Hence when such task occurs, several adjacent rows or columns of the cache elements may be written to the fuzzy buffer and re-initialized even though some of the elements may not need to be re-claimed.
Direct Computation of the Fuzzy Region of an Edge
Assume that the equation of the edge is expressed as:
x=az+b (1)
y=cz+d (2)
2. Assume that all the coordinates of a viewpoint P in the current viewpoint bounding box is (X0,Y0,Z0).
3. The image projection of a point T (x,y,z) on the edge with respect to P is ##EQU2##
4. Since (x,y,z) and (x0,y0,z0) are uncertain, the region comprising all possible values of (X,Y) forms an area which is the fuzzy region of the edge.
5. By substituting (2) into (4) and rearranging, ##EQU3##
If Y is known, the extrema of z can be found by substituting the appropriate extrema of y0 and z0 into the above equation. All points on the edge having z within these extrema could project at Y from viewpoints in the current viewpoint bounding box.
6. substituting (1) into (3), ##EQU4##
Notice that in (6) there is no local extremum of X for changing z. Hence by assigning the extrema of z found from section 5 and the appropriate extreme values of x0 and z0 into (6), the two extrema of X can be found. These values correspond to the maximum and minimum possible X for a particular value of Y.
7. A scan line approach can be used to approximate the fuzzy region of the edge. Substituting (2) into (4) and rearranging, ##EQU5##
By assigning the appropriate extreme values of z, y0 and z0 into (7), the maximum and minimum values of Y are obtained. These values are rounded up and down respectively to cover all scan lines the edge may project onto. Each of these scan lines corresponds to a value of Y. From section 5 and 6, the maximum and minimum of X can be found. The area within these values is the scan line segment of the fuzzy region of the edge. The corresponding segment of the expanded fuzzy region of that edge is the further rounding of the segment to fully cover all the grid cells it projects cover.
The above technique computes the expanded fuzzy regions by scan-converting the segments of the regions on the X scan lines. For edges more horizontal than vertical, the expanded fuzzy regions may be more accurately computed by scan-converting the segments of the regions on the Y scan lines.
This is shown in FIG. 14 where the fuzzy region takes on a scan-line shape about the fuzzy region of the edge. This is substantially smaller than the fuzzy extent of the edge.
SCAN-CONVERSION OF PATCHES ON THE HEMICUBE PYRAMID
This appendix describes an efficient technique for scan-converting a patch on a plane which contains cells with different resolutions. It can be used to compute the form-factor of patches in the radiosity method with or without the use of the fuzzy projection techniques.
The technique uses a data structure called the hemicube pyramid. Each level of the pyramid contains a set of 2D arrays. Each array corresponds to a level and a projection face. Each element of the array corresponds to a pixel on the plane. Note that the hemicube may be a prism rather than a cube and not all its planes need to be active.
An element in the array is called a pyramid element. In addition to the information needed in an original hemicube element, it contains an sub-division indicator and a pointer. The sub-division indicator is initially set to OFF. The indicator becomes active if its level is less than the required accuracy level of patches projected onto it.
The pointer is used to link a pyramid element with one of the pyramid elements at the higher accuracy level and within its region. To enable more accurate sampling, the elements to be linked are usually chosen such that an element accessed by the pointers from an element at the lower accuracy level is as close to the center of the latter as possible.
The organizaton of the hemicube pyramid corresponding to a face of the hemicube is shown in FIG. 15.
After the initialization of the hemicube pyramid, patches are accessed in decreasing order of accuracy and scan-converted to the hemicube pyramid. This order of access ensures that patches are scan-converted in resolutions equal to or finer than their own optimal projection resolutions and the optimal projection resolutions of patches they might occlude.
For each projecting patch, the hemi-cube projection resolution corresponding to its accuracy level is found. Each hemicube pyramid element belonging to that resolution and under the projection of that patch is accessed.
If the accessed element is already at the most accurate level, or if its sub-division indicator is not on, the ID of the current patch is written into the current element if that patch is closer than the so-far closest patch stored in the element, as in the normal hemicube scan-conversion. If this update occurs, all the ancestors of the current element until the element whose sub-division indicator has already been set to on are accessed. The sub-division indicator of each accessed element is set to on.
Note that the recursive access of ancestor elements can instead be carried out for all elements under the patch projection immediately after they have been found. However, if the possibility that patches are invisible is high, carrying out the recursion after determining that the patch is closer is more efficient as it does not need to be applied on elements if that patch is hidden.
If the sub-division indicator of the current element is ON, patches requiring more accurate form-factor evaluation have been projected onto the element. Therefore, each of its daughter elements under the projection of the current patch is accessed and the above computations are repeated.
After the scan-conversion of all projecting patches, the elements in all the levels are scanned. From the projection information stored in these elements, the form factors of the patches are computed.
If highly accurate form-factor computations and hence very high hemi-cube resolutions are required, relatively large amounts of memory are needed to store the hemi-cube arrays. The initialization and scanning of these arrays for form-factor accumulation is also costly.
Since patches requiring very fine projections are relatively few, the above problem can be overcome by using sparse matrices to store active (sub-divided) elements corresponding to high resolution levels. Each matrix is a one-dimensional array. Active elements are stored in it at positions determined by a hashing function parameterized by pixel positions.
High accuracy form-factor computations can also be achieved using ray-tracing in conjunction with hemicube computations. All the patches are still scan-converted to the hemicube pyramid or the normal hemicube. However, the scan-conversion of patches requiring high accuracy in
REFERENCES
1. Aggarwal J. K., "Dynamic Scene Analysis," in "Image Sequence Processing and Dynamic Scene Analysis," Huang, T. S. (Eds,) 1981, Springer-Verlag, pp. 40-73.
2. Appel A. "The Notion of Quantitative Invisibility and the Machine Rendering of Solids", Proc. of the ACM National Conference, 1967, Thompson Books, Washington D.C., pp.387-393.
3. Baum, D. R., Rushmeier H. E., Winget J. M., "Improving Radiosity Solutions through the use of Analytically Determined Form-factors", Computer Graphics, Vol 23 Number 3, 1989, pp. 325-334.
4. Bradler, N., Tsotsos, J. K., (Eds) "Motion: Representation and Perception," North-Holland, 1986.
5. Catmull, E. "A Subdivision Algorithm for Computer Display of Curved Surfaces", PhD Thesis, Report UTEC-CSc-74, Universitv of Utah, 1975.
6. Chang, S. S. L., Zadeh, L. A., "On Fuzzy Mapping and Control," IEEE Transactions on Systems, Man, and Cybernatics, Vol.SMC-2, No. 1, January 1972, pp. 30-34.
7. Cohen, M. F., D. P. Greenberg, "The Hemi-Cube: A Radiosity Solution for Complex Environment," SIGGRAPH 85, pg. 31-40.
8. Cohen, M. F., Chen, S. E., Wallace, J. R., Greenberg, D. P., "A Progressive Refinement Approach to Fast Radiosity Image Generation," SIGGRAPH 88, pg. 75-84.
9. Crocker, G. A., "Invisibility Coherence for Faster Scan-Line Hidden Surface Algorithms," SIGGRAPH'84, pp. 315-321.
10. Fournier, A., Fussell, D., "On the power of the Frame Buffer,", ACM Transactions of Graphics, April 1988, pp. 103-128.
11. Fuch, H., Kedem, Z., Naylor, B. F., "On Visible Surface Generation by A Priority Tree Structures," SIGGRAPH'80, pp. 124-133.
12. Gordon, D., "Front-to-Back Display of BSP Trees," IEEE CG&A, September 1991, pp. 79-85.
13. Haines, E. A., Wallace, J. R., "Shaft Culling for Efficient Ray-Traced Radiosity," unpublished paper, July 1991.
14. Hornung, C. "A Method for Solving the Visibility Problem", IEEE Computer GRaphics and Applications, July 1984, pp.26-33

Claims

1. A method of reducing the complexity of visibility calculations required for the production of multi-dimensional computer generated images, said method performed on a computer, said method comprising the steps of:
prior to an occlusion or invisibility relationship computation being carried out on a plurality of surfaces from selected viewpoints to be calculated, said viewpoints are divided into groups;
for selected ones of said surfaces, determining for each said group whether each said selected surface is
(a) an always occluded surface, an always hidden surface, or a remaining surface; or
(b) an always unoccluded surface, or a remaining surface; or
(c) an always hidden surface, or a remaining surface;
wherein said remaining surface is a surface which is unable to be determined with certainty as to whether it is either unoccluded or hidden;
exempting from said occlusion or invisibility relationship computation those surfaces which are either always unoccluded or always hidden;
maintaining a record of said remaining surfaces; and
carrying out occlusion or invisibility relationship computations on said remaining surfaces;
whereby the production of multi-dimensional computer generated images is facilitated.
2. A method as claimed in claim 1, wherein the maintaining step comprises updating a previous record.
3. A method as claimed in claim 1, wherein the maintaining step comprises creating a new record.
4. A method as claimed in claim 1, wherein said images are selected from the group consisting of graphic images, computer vision data, abstract data and physical data.
5. A method as claimed in claim 1, wherein the reduction in complexity involves a reduction in the number and/or duration of visibility calculations.
6. A method as claimed in claim 1, wherein the visibility calculations form part of a series of radiosity calculations.
7. A method as claimed in claim 6, wherein the exempting step involves the direct computation of said unoccluded surfaces.
8. A method as claimed in claim 1, wherein one or more of said surfaces are sub-sets of a primary surface.
9. A method as claimed in claim 1, further comprising the following steps performed with respect to selected ones of said surfaces:
(6) determining the combined projection of said selected ones of said surfaces and determining the combined projection of the boundary edges of said each selected surface;
(7) subtracting from the combined surface projection the combined boundary edge projection to form an umbra region of said each selected surface; and
(8) carrying out said occlusion or invisibility relationship computations on said umbra regions and said combined projection of said each selected surface.
10. A method as claimed in claim 9, wherein the extent of the boundary area is coincident with a scan-line representation of any of said selected surfaces.
11. A method as claimed in claim 9 or 10, wherein the combined projection of said selected ones of said surfaces is replaced with a projection area of the selected ones of said surfaces from a viewpoint in said group.
12. A method as claimed in claim 11, wherein said calculations are further simplified by detecting, and not carrying out an occlusion or invisibility relationship computation for, surface areas which are totally hidden as a consequence of their depth from the current group of viewpoints and the presence of nearby surface areas of lesser depth when viewed from the current group of viewpoints, said detecting of said totally hidden surface areas comprising the steps of:
for each range of viewing directions related to said group of viewpoints allocating a depth plane to the projection surface of the corresponding surfaces in that viewing direction, each element of the projection surface having a plurality of depth values associated with the viewpoints in the group viewing the corresponding surface area;
for each surface area selecting the depth value for each element and storing same to define the depth values of the projection surface, thus representing the depth value of the umbra region;
similarly determining the depth value of the combined surface projection; and
comparing the depth value of the umbra region with that of the combined surface projection, and if the former is smaller, then deeming that surface forming the combined surface projection is totally hidden.
13. A method as claimed in claim 12 wherein the selecting step selects the smallest of the approximate largest depth values of surfaces projected on each element.
14. A method as claimed in claim 12 wherein the determining step selects the smallest possible depth value of each element.
15. A method as claimed in claim 1, wherein said calculations are further simplified by detecting, and not carrying out an occlusion or invisibility relationship computation for, surface areas which are totally unoccluded, said detecting of said totally unoccluded areas comprising the steps of:
prior to the carry-out step, creating and storing a combined projection for each of selected ones of said surfaces or surface elements and for each surface or element of each stored combined projection, indicating whether the surface or element is either;
(a) not within the combined projection of any of said selected surfaces;
(b) within the combined projection of a single one of said selected surfaces; or
(c) within the combined projection of more than one of said selected surfaces; and
scanning the combined projection of each said selected surface and where, in all elements for one projection are indicated by step (14(b)), deeming the surface so projected to be totally unoccluded.
16. A method as claimed in claim 1, wherein said surfaces are arranged in a hierarchy represented by varying levels of details.
17. A method of reducing the visibility related computations in an environment consisting of three-dimensional abstract or physical surfaces, said method performed on a computer, said method comprising the steps of:
prior to a visibility computation being carried out, some or all viewpoints or possible occurrences of viewpoints are classified into groups of viewpoints;
defining one or more projection surfaces for the purpose of generating simultaneous projections of surfaces or surface elements with respect to a group of viewpoints;
for selected surfaces, and for each group of viewpoints, determining whether those surfaces or their surface elements are always invisible to said group by computing and comparing their projection mappings or simultaneous projections on said projection surfaces;
ignoring or treating specially some or all surfaces or surface elements which are always invisible to said group of viewpoints during the actual visibility or visibility related computations for some or all viewpoints in said groups;
whereby the visibility related computations are reduced.
18. A method of reducing the visibility, radiosity or visibility related computations in an environment consisting of three-dimensional abstract or physical surfaces, said method performed on a computer, said method comprising the steps of:
prior to a visibility computation being carried out, classifying some or all viewpoints or possible occurrences of viewpoints into groups of viewpoints;
defining one or more projection surfaces for the purpose of the simultaneous projections of surfaces or surface elements with respect to a group of viewpoints;
dividing each of said projection surfaces into regular or irregular grids;
defining a data structure which organizes computer storage for storing of projections on said grids;
for each of the selected surfaces or surface elements and for each group of viewpoints, simultaneously projecting said surfaces or their surface elements onto the projection surfaces, computing grid cells which are under the projections; and
ignoring or treating specially surface elements which are always visible to said group of viewpoints during the actual visibility or visibility related computations for selected viewpoints in said group;
whereby the visibility, radiosity or visibility related computations are reduced.
19. A method of storing computer generated image data, said method performed on a computer having a physical data store, said method comprising the steps of:
identifying a plurality of image portions visible to a selected group of viewpoints;
selecting those portions having a predetermined image characteristic and storing same in an organization storage structure in the physical data store;
for a first one of said selected portions, allocating a portion identifier and storing same together with image data of said first portion in the physical data store; and
for the remaining said selected portions, calculating a difference value identifier and storing same together corresponding image data of said selected portion, wherein any said selected portion is addressable by means of said portion identifier and one or more said difference value identifiers.
20. A method as claimed in claim 19 wherein the calculated difference value identifier represents an address difference value between a said remaining selected portion and said first selected portion.
21. A method as claimed in claim 19 wherein the calculated difference value identifier represents an address difference value between a said remaining selected portion and another said remaining selected portion, in which one of said remaining selected portions has a calculated difference value identifier that represents an address difference value between a said remaining selected portion and said first selected portion.
22. A method as claimed in claim 19 wherein said storage structure is non-contiguous.
23. A method as claimed in claim 22 wherein said storage structure is a linked-list.
24. A method as claimed in claim 19 wherein said storage structure is contiguous.
25. A method as claimed in claim 22 wherein said storage structure is an array.
26. A method as claimed in claim 19 wherein said image portions are stored based upon a plurality of said predetermined image characteristics to give multiples sets of image data, in which any element of any set can be manipulated based upon a viewing from another group of viewpoints.
27. A method as claimed in claim 26 wherein said manipulation involves adding an element to a set, deleting an element from a set, or copying an element between any two or more sets.
28. A method as claimed in claim 27 wherein a plurality said predetermined image characteristics are available and only the smallest of the plurality less one corresponding set(s) are(is) retained, the other one being redundant.
29. A method of reducing the complexity of calculations required for the production of multi-dimensional computer generated images or the reduction of multi-dimensional data to multi-dimensional data having at least one less dimension, said method performed on a computer, said method comprising the steps of:
observing an object surface with a plurality of viewpoints, and defining a projection surface for each said viewpoint;
creating a working surface and mapping each of said projection surfaces onto said working surface;
computing a combined region on said working surface that contains at least one of the projection surfaces;
whereby the complexity of calculations required for the production of multi-dimensional computer generated images or the reduction of multi-dimensional data to multi-dimensional data having at least one less dimension is reduced.
30. A method as claimed in claim 29 wherein prior to the computing step occurring, the step of:
approximating the viewpoints by calculating a volume containing all of the viewpoints; and
in which the computing step uses said volume to approximate all said viewpoints.
31. A method of reducing the complexity of calculations required for the production of multi-dimensional computer generated images or the reduction of multi-dimensional data to multi-dimensional data having at least one less dimension, said method performed on a computer, said method comprising the steps of:
observing an object surface with a plurality of viewpoints, and defining a projection surface for each said viewpoint;
creating a working surface and mapping each of said projection surfaces onto said working surface;
computing a region that contains all the mappings;
whereby the complexity of calculations required for the production of multi-dimensional computer generated images or the reduction of multi-dimensional data to multi-dimensional data having at least one less dimension is reduced.
32. A method as claimed in claim 31 wherein prior to the computing step occurring, the step of:
approximating the viewpoints by calculating a volume containing all of the viewpoints; and
in which the computing step uses said volume to approximate all said viewpoints.
33. A method as claimed in claim 12 wherein said calculations for said remaining surfaces are further reduced in complexity by:
establishing a plurality of accuracy levels required to be reached by form-factor computations of said remaining surfaces;
allocating one of said accuracy levels to each of said always occluded surfaces, to each of said always hidden surfaces, and to each element of said combined projection;
updating the accuracy level of each said surface with the accuracy level of the corresponding element of the combined projection if the latter is larger, and subsequently scan-converting each of said surfaces in order commencing with those having the highest accuracy level and ending with those having the lowest accuracy level, the updated accuracy levels indicating the maximum level of accuracy for form-factor computations for each said surface; and
determining if the updated accuracy level for each said surface is higher than that which said surface can provide, and if so, recursively subdividing the corresponding said surface into a plurality of sub-surfaces until the corresponding accuracy levels of said sub-surfaces match that of said surface.
34. A method as claimed in claim 33 wherein said form-factor computations are performed using a ray-tracing technique or a hemi-cube technique.
35. A method of scan-converting a plurality of surfaces of a multi-dimensional computer generated image, said method performed on a computer, said method comprising the steps of:
creating a data structure comprising a plurality of two-dimensional arrays, each said array relating to a corresponding accuracy level for a projection of said surfaces as viewed from a viewpoint, the elements of each array being pixels on a projection plane;
for each said element, providing an associated sub-division indicator, the sub-division indicator being adapted to indicate an active sub-division of the corresponding said element;
initializing each said sub-division indicator to OFF;
for each patch of said surface, defining a desired accuracy for same and scan-converting the projection including said patches in decreasing order of accuracy into corresponding elements of said structure, and setting the sub-division indicator for each said element of the highest accuracy level to ON if that element corresponds to one of said patches;
for each said element of the surface, if the sub-division indicator is ON, accessing the sub-elements of said element which are under the projection, and of the indicator is not ON, updating the patch in form to said sub-element, if the depth of the patch is less than the depth stored in the sub-element;
accessing each said element and adding a form-factor contribution of the element to the form-factor of a record of the patch whose own identifier is stored in the patch;
whereby a plurality of surfaces of a multi-dimensional computer generated image are scan-converted.

Priority Claims (5)

Number	Date	Country
PK6942	Jun 1991	AUX
PK7305	Jul 1991	AUX
PK8643	Oct 1991	AUX
PK8645	Oct 1991	AUX
PK9218	Oct 1991	AUX

PCT Information

Filing Document	Filing Date	Country	Kind	102e Date	371c Date
PCT/AU92/00302	6/19/1992			6/1/1994	6/1/1994

Publishing Document	Publishing Date	Country	Kind
WO93/00650	1/7/1993

US Referenced Citations (26)

Number	Name	Date
4594673	Holly	Jun 1986
4625289	Rockwood	Nov 1986
4697178	Heckel	Sep 1987
4819192	Kuragano et al.	Apr 1989
4825391	Merz	Apr 1989
4855938	Gonzalez-Lopez et al.	Aug 1989
4901252	Fitzgerald et al.	Feb 1990
4918626	Watkins et al.	Apr 1990
4928250	Greenberg et al.	May 1990
5027292	Matsumoto	Jun 1991
5058042	Hanna et al.	Oct 1991
5081698	Kohn	Jan 1992
5084830	Doornink et al.	Jan 1992
5086496	Mulmuley	Feb 1992
5088054	Paris, II	Feb 1992
5159663	Fossum	Oct 1992
5253335	Mochizuki et al.	Oct 1993
5268996	Steiner et al.	Dec 1993
5295243	Robertson et al.	Mar 1994
5299298	Elmquist et al.	Mar 1994
5313568	Wallace et al.	May 1994
5377313	Scheibl	Dec 1994
5402532	Epstein et al.	Mar 1995
5414801	Smith et al.	May 1995
5448686	Borrel et al.	Sep 1995
5619593	Ono	Apr 1997

Foreign Referenced Citations (3)

Number	Date	Country
193151	Sep 1986	EPX
0 481 581	Apr 1992	EPX
2 228 850	Sep 1990	GBX

Non-Patent Literature Citations (214)

Entry
Stabbing Isothetic Boxes & Rectangles in 0(n lg n) time M. Hohmeyer S. J. Teller 1991.
A Characterization of Ten Hidden-Surface Algorithms I. E. Sutherland, R. F. Sproull, R. A. Schumacker Computing Surveys, vol. 6, No. 1, 1975.
Fast Algorithms for 3D-Graphics G. Glaeser, 1994.
Imprecision in Computer Vision R. Jain, S. Haynes Advances in Fuzzy Sets, Possibility and Applications, 1983.
Accelerated radiosity method for complex environments H. Xu Q. Peng, Y. Liang Eurographics'89 1989.
Application Challenges to Computational Geometry, CG Impact Task Force Report CG Impact Task Force Technical Report TR-521-96, Princeton University 1996.
Computer Graphics, Principles and Practice, 2nd edition J. D. Foley A. van Dam S. K. Feiner J. F. Hughes.
New Trends in Animation and Visualization N. Thalmann D. Thalmann 1991.
Graphics Systems: Architecture & Realization R. Andreev 1993.
Analysis of Radiosity Techniques in Computer Graphics B. Kwok MSc Thesis York University, May 1992.
Image display data computer forming method--uses perspective transformation with calculation time reduction on shadow and hidden surface processing Sony 86.233751/36.
Stabbing and ray shooting in 3 dimensional space M. Pellegrini 1990.
The Geometry of Beam Tracing N. Dadoun D. Kirkpatrick 1985.
Algorithms for line transversals in space D. Avis CG1987.
Optimization of the binary space partition algorithm (BSP) for the visualization of dynamic scenes E. Torres Eurographics'90.
A mathematical Semantics of Rendering II. Approximation E. Fiume CVGIP: Graphical Models and Image Processing vol. 53, No. 1, Jan. 1991.
A Characterization of Ten Rasterization Techniques N. Gharachorloo S. Gupta R. Sproull I. Sutherland Computer Graphics, vol. 23, No. 3 (Siggraph'89) 1989.
The use of projective geometry in Computer Graphics I. Herman 1992.
Ray Tracing with Cones J. Amanatides Computer Graphics, vol. 18, No. 3 (Siggraph'84) 1984.
The A-buffer, an Antialiased hidden surface method L. Carpenter Computer Graphics, vol. 18, No. 3, (Siggraph'84) 1984.
Principles and Applications of Pencil Tracing M. Shinya T. Takahashi S. Naito Computer Graphics, vol. 21, No. 4, (Siggraph'87) 1987.
Light-water interaction using backward beam tracing M. Watt Computer Graphics, vol. 24, No. 4 (Siggraph'90) 1990.
Rendering CSG Model with a ZZ-buffer D. Salesin J. Stolfi Computer Graphics, vol. 24, No. 4 (Siggraph'90) 1990.
A solution to the hidden-line problem for computer-drawn polyhedra P. Loutrel IEEE Transactions on Computers, Mar. 1970.
Sorting and the hidden-surface problem I. Sutherland R. Sproull R. Schumacker National Computer Conference, 1970.
The Notion of quantitative invisibility and the machine rendering of solids A Appel ACM National Meeting 1967.
An analytic visible surface algorithm for independent pixel processing E. Catmull Computer Graphics, vol. 18. No. 3 (Siggraph'84) 1984.
Computing the lines piercing four lines S. Teller M. Hohmeyer Technical Report 92-665, University of California, Berkeley.
Computing the antipenumbra of an area light source S. Teller Computer Graphics, vol. 26, No. 2, (Siggraph'92) 1992.
Computer animation, theory and practice, second revised edition N. Thalmann D. Thalmann 1990.
Image Synthesis M. Brest 1992.
Management of large amounts of data in interactive building walkthroughs T. Funkhouser C. Sequin S. Teller Symposium on Interactive 3D Graphics, 1992.
Temporal Coherence in Ray Tracing S. H. Badt PhD thesis, University of Texas at Dallas, 1989.
Near real-time shadow generation using BSP trees N. Chin S. Feiner Computer Graphics, vol. 23, No. 3 (Siggraph'89) 1989.
Adaptive Display algorithm for interactive frame rates during visualization of complex virtual environments T. Funkhouser C. H. Sequin Siggraph'93 1993.
Modeling global diffuse illumination for image synthesis A. Campbell, III. PhD Thesis University of Texas at Austin 1991.
A 3-dimensional representation for fast rendering of complex scenes S. Rubin T. Whitted 1980.
Radiosity redistribution for dynamic environment D. George F. Sillion D. Greenberg IEEE Computer Graphics & Applications, vol. 4, 1990.
A survey of shadow algorithms A. Woo, F. Poulin A. Fournier IEEE Computer Graphics & Applications, vol. 6, 1990.
Error-bounded antialiased rendering of complex environments N. Greene M. Kass Siggraph'94 1994.
Fast computation of shadow boundaries using spatial coherence and backprojections A. Stewart S. Ghali Siggraph'94 1994.
A fast shadow algorithm for area light sources using backprojection G. Drettakis E. Fiume Siggraph'94 1994.
Stabbing oriented convex polygons in randomized O(n 2) time. S. Teller M. Hohmeyer Contemporary Mathematics, 1994.
Obscuration culling on parallel graphics architecture C. George Technical report TR95-017 Univerisity of North Carolina at Chapel Hill 1995.
Visibility between two edges of a simple polygon D. Avis T. Gum G. Goussaint The Visual Computer, 1986, No. 2.
Increasing update rates in the building walkthrough system with automatic model-space subdivision and potentially visible set calculations J. M. Airey PhD Thesis, University of North Carolina, 1990.
Realism in computer graphics: a survey. J.Amanatides IEEE Computer Graphics and Applications, vol. 7, No. 1, 1987.
Finding a line transversal of axis objects in three dimensions N. Amenta Proc. 3rd Annual ACM-SIAM Symposium on Discrete Algorithms, 1992.
Beyong the third dimension, geometry, computer graphics and higher dimension Banchoff T. F. Scientific American Library 1990.
A general version of crow's shadow volumes Bergeron P. IEEE Computer Graphics and Applications, vol. 6, No. 9, Sep. 1986.
Me and my (fake shadow) Blinn, J. F. IEEE Computer Graphics and Applications, vol. 8, No. 1 1988.
Generating soft shadows with a depth buffer algorithm Brotman L. S. IEEE Computer Graphics and Applications vol. 4, No. 10, 1984.
A multiresolution spline with application to image mosaics Burt P. J. Adelson E. H. ACM Transactions on Graphics, vol. 2, Oct. 1983.
Hierarchical geometric models for visible surface algorithms Clark, J. H. Communications of the ACM, vol. 19, No. 10, Oct. 1976.
An overview of rendering techniques Dennis A. R. Computer & Graphics, vol. 14, No. 1, 1990.
Hybrid shadow testing scheme for ray tracing Eo K. S. Kyung C. M. Computer Aided Design vol. 21, No. 1 Jan. 1989.
Hierarchical rendering of complex environments Greene, N. PhD dissertation, University of California, Santa Cruz 1995.
Bibliography of hidden-line and hidden-surface algorithms Griffiths, J. G. Computer-Aided Design, vol. 10, May 1978.
The light buffer: a shadow-testing accelerator Haines, E. A. Greenberg, D. IEEE Computer Graphics and Applications, vol. 6, No. 9, Sep. 1986.
Algorithms for antialiased cast shadows Hourcade J. C. Nicolas A. Computer and Graphics, vol. 9, No. 3 1985.
Hemi-cube ray-tracing: a method for generating soft shadows Meyer U. Proceedings, Eurographics'90, 1990.
Principles of interactive computer graphics, 2nd edition Newman, W. M. Sproull, R. F. 1979.
Principles of interactive computer graphics, 1st edition Newman, W. M. Sproull, R. F. 1973.
A comparison of four visibility acceleration techniques for radiosity Ng, A. The visual computer, 1996, vol. 12, pp. 307-316.
Shading models for point and linear sources Nishita, T. Okamura, I. Nakamae, E. ACM transactions on graphics, vol. 4, No. 2, Apr. 1985.
Radiosity of dynamic scenes in flatland with the visibility complex Orti, R. Riviere S. Durand F. Puech C. Proceedings, Eurographics'96 1996.
Visibility, occlusion, and the aspect graph Plantinga H. Dyer C. R. International Journal of computer vision, vol. 5, No. 2, 1990.
Necessary and sufficient conditions for hyperplane transversal Pollack R. Wenger R. Proc. 5th Annual Symposium on Computational Geometry, 1989.
Shading and shadowing with linear light sources Poulin P. Amanatides J. Proceedings, Eurographics'90 1990.
Rendering antialiased shadows with depth maps Reeves W. Salesin D. Cook R. Proceedings, Siggraph'87 1987.
Machine perception of three-dimensional solids Roberts L. G. Optical and Electro-Optical Information Processing, J. Tippett (editor) MIT Press, 1965.
Procedureal elements for computer graphics Rogers D. F. McGraw-Hill(publisher) 1985.
Optics B. Rossi Addison-Wesley(publisher) 1957.
An optimal algorithm for detecting weak visibility of a polygon J. Sack S. Suri IEEE transactions on computers, vol. 39, No. 10, Oct. 1990.
Shaded rendering and shadow computation for polyhedral animation Seales W. B. Dyer C. R. Proceedings, Graphics Interface '90, May 1990.
Optics, third edition Sears F. W. Addison-Wesley Publishing Co. 1949.
Linear programming and convex hulls made easy Seidel R. Proc. 6th ACM Symposium on COmputational Geometry, 1990.
Output-sensitive visibility algorithms for dynamic scenes with applications to virtual reality Sudarsky O. Gotsman C. Proceedings, Eurographics'96 1996.
Automatic view function generation for walk-through animation using a reeb graph Shinagawa Y. kunii T. Nomura Y. Okuna T. Young Y. Computer Animation '90 1990.
Octant priority for radiosity image rendering Wang Y. Davis W. Proceedings, Graphics Interface'90 1990.
Casting curved shadows on curved surfaces Williams L. Proceedings, Siggraph'78 1978.
Pyramidal parametrics William L. Proceedings, Siggraph'83 1983.
Accelerated radiosity method for complex environments Xu H. Peng Q. Liang Y. Proceedings, Eurographics'89 1989.
Pyramid clipping for efficient ray traversal Zwann M. Reinhard E. Jansen F. Proceedings of the Sixth Eurographics Rendering Workshop 1995.
Application challenges to computational geometry, CG impact task force report CG impact task force Princeton University Computer Science Dept. Technical Report TR-521-96 http://graphics.lcs.mit.edu/.about.seth/pubs/taskforce/techrep.html http://www.cs.princeton.edu/.about.chazelle/taskforce/CGreport.ps.Z http://graphics.lcs.mit.edu/.about.seth/pubs/pubs.html 1996.
My response to application challenges to computational geometry Franklin R. http://www.ecse.rpi.edu/Homepages/wrf/geom.sub.-- response.html http://netlib.bell-labs.com/netlib/compgeom/discuss/archive/96/ta skforce.html 1996.
Comments on the report "Application challenges to computational geometry" Heckbert P. http://www.cs.duke.edu/.about.jeffe/compgeom/files/heckbert.html http://netlib.bell-labs.com/netlib/compgeom/discuss/archive/96/ta skforce.html 1996.
Follow-up comments on the report "Application challenges to computational geometry" Coulson T. http://www.cs.duke.edu/.about.jeffe/compgeom/files/coulson.html http://netlib.bell-labs.com/netlib/compgeom/discuss/archive/96/ta skforce.html 1996.
Inside Quake: visible surface determination http://www.gamers.org/dEngine/quake/papers/ddjpvs.html 1996.
CGDC Quake Talk http://www.gamers.org/dEngine/quake/papers/mikeab-cgdc.html 1996.
Quake hidden surface removal http://www.gamers.org/dEngine/quake/papers/ddjzsort.html 1996.
Quake editing tools information http://www.gamers.org/dEngine/quake/QuakeEd/qedit.sub.-- infor.html 1996.
Zen of graphics programming Abrash M. 1996.
Computer graphics: more unsolved problems Siggraph'91 panel 1991.
Global illumination in architecture and entertainment Siggraph'96 course notes 1996.
Interactive walkthrough of large geometric databases Siggraph'96 course notes 1996.
Imprecise computation and load sharing in computer generated imaging system Berger M. Zhao W. Graphics Interface '90 1990.
Exploiting temporal coherence in ray tracing Chapman J. Calvert T. Sill J. Graphics Interface'90 1990.
Approximate ray tracing Dauenhauer D. Graphics Interface '90 1990.
Approximate and probabilistic algorithms for shading and rendering structured particle systems Reeves W. Siggraph 1985 1985.
Fundamentals of interactive computer graphics Foley J.D. van Dam A. Addison-Wesley Publishing Co. 1982.
Mulit-pass multi-resolution algorithm for the determination of hidden surfaces Lim, H. L. Inter-faculty symposium on computer graphics and image processing 1987.
Ten Unsolved Problems in Rendering Heckbert P. S. Workshop on Rendering Algorithms and Systems, Gaphics Interface'87 www addr: http://www.cs.cmu/.about.ph (index page) 1987.
cainter's & z-buffer algorithms Course note, CS deprt, Carnegie-Melon university www addr: http://www.cs.cmu.edu/afs/cs/project/anim/ph/463.95/pub/www/notes.toc.html 1996.
3D comes alive Ozer, J. PC magazine, Jun. 25, 1996.
Affordable 3-D workstations Hummel, R. L. Byte Dec. 1996.
Gaming in the next dimension Case, L. Salvator, D. Computer Gaming Jul. 1996.
3D engine list Isakovic, K. www addr: http://www.cs.tu-berlin.de/.about.ki/engines.html 1996.
OpenGL the leading visual programming interface www addr: http://www.sgi.com/Products/Dev.sub.-- environ.sub.-- ds.html 1996.
3D computer graphics (second edition) Glassner A. S. Design Press 1989
The UC Berkeley system for interactive visualization of large architectural models T. Funkhouser S. Teller C. Sequin D. Khorramabadi Presence, vol. 5, No. 1, Winter 1996.
Visibility computation for efficient walkthrough of complex environment R. Yagel R. William Presence, vol. 5, No. 1, Winter 1996.
Large models for virtual environments: A review of work by the architecutal walkthrough project at UNC M. R. Mine H. Weber Presence, vol. 5, No. 1, Winter 1996.
A hidden-line algorithm for hyperspace R. P. Burton D. R. Smith SIAM Journal of Computing vol. 11, No. 1, Feb. 1982.
Canonic representations for the geometrics of multiple projective views Q. T. Luong T. Vieville University of California at Berkely CS Dept. Tech. Report UCB/CSD093-772 1993.
Multidimensional graphing in two-dimensional spaces T. Mihalisin E. Gawlinski J. Timlin J. Schwegler Computers in Physics, Nov./Dec. 1989.
A framework for global illumination in animated environments J. Mineroff J. Dorsey H. Rushmeier Rendering Techniques'95 (Proceedings of the Eurographics Workshop in Dublin, Ireland Jun. 1995.
Shadows for bump-mapped surfaces N. L. Max Advanced Computer Graphics (Proceedings of Computer Graphics Tokyo '86) 1986.
The simulatoin of natural features using cone tracing D. Kirk Advanced Computer Graphics (Proceedings of Computer Graphics Tokyo '86) 1986.
Simulating soft shadows with graphics hardware P. S. Heckbert M. Herf Carnegie-Melon University CS Dept. Technical Report CMU-CS-97-104 1997.
Spatial transformation for rapid scan-line surface shadowing P. Robertson IEEE computer graphics & applications Mar. 1989.
Incremental update of the visibility map as seen by a moving viewpoint in two dimensions S. Ghali A. J. Stewart Eurographics Workshop on Animation and Simulation, Aug. 1996.
Z. H. Zhao D. Dobkin Continuous algorithms for visibility: the space searching approach fourth eurographics workshop on rendering 1993.
Y. Shinagawa S. Miyoshi T. Kunii Viewpoint analysis of drawings and paintings rendered using multiple viewpoints: cases containing rectangular objects Fourth Eurographics Workshop on Rendering 1993.
F. Jansen A. Chalmers Realism in real time Fourth Eurographcis Workshop on Rendering 1993.
D.R. Baum et al., "The back-buffer algorithm: an extension of the radiosity method to dynamic environments", The Visual Computer, vol. 2, No. 5, pp. 298-306, 1986.
S.J. Teller er al., "Visibility preprocessing for interactive walkthroughs", Computer Graphics, vol. 25, No. 4, pp. 61-69, 1991.
J. Marks et al., "Image and intervisibility coherence in rendering", Graphics Interface '90 Proceedings, pp. 17-30, 1990.
H.L. Lim, "Rendering techniques in three-dimensional computer graphics", Ph.D. thesis, University of Sydney, 1993.
S.J. Teller, "Visibility computations in densely occluded polyhedral environment", Ph.D. thesis, University of California at Berkeley, 1992.
A.S. Glassner, "Principles of Digital Image Synthesis", vol. 2, Morgan Kaufmann Publisher, San Francisco, 1995.
D. Gordon et al., "Front-to-back display of BSP trees", IEEE Computer Graphics & Applications, vol. 11, pp. 79-85, 1991.
K.L. Shelley et al., "Path specification and path coherence", Computer Graphics, vol. 16, No. 3, pp. 157-161, 1982.
J. Vilaplana et al., "Exploiting coherence for clipping and view transformations in radiosity algorithms", in Eurographics Workshop on Photosimulation, Realism and Physics in Computer Graphics, Rennes, France, pp. 137-149, 1989.
C.B. Jones, "A new approach to the `hidden line` problem", Computer Journal, vol. 14, No. 3, pp. 232-236, 1971.
F.C. Crow, "Shadow algorithms for computer graphics", Computer Graphics, vol. 11, No. 2, pp. 442-448, 1977.
X. Pueyo, "The use of visibility coherence for radiosity computation", in First International Conference on Visualization and Intelligent Design in Engineering and Architecture, pp. 17-28, 1993.
T. Nishita et al., "Continuous tone representation of three-dimensional objects taking account of shadows and interreflection", Computer Graphics, vol. 19, No. 3, pp. 23-30, 1985.
T. Funkhouser, "Database and display algorithms for interactive visualization of architectural models", Ph.D. thesis, University of California at Berkeley, 1993.
S.J. Teller et al., "Global visibility algorithms for illumination computations", in Computer Graphics (Siggraph '93), vol. 27, pp. 239-246, 1993.
S. Coorg et al., "Temporally coherent conservative visibility", in Twelfth Annual ACM Symposium on Computational Geometry, Philadelphia, ACM Press, New York, pp. 1-10, 1996.
S. Coorg et al., "A spatially and temporally coherent object space visibility algorithm", Laboratory of Computer Science, MIT, Technical Report TM-546, 1996.
E. Catmull, "Computer display of curved surfaces" in IEEE Transactions of Computers, pp. 309-315, 1971.
M.F. Cohen et al., "The hemi-cube: A Radiosity solution for complex environment", Computer Graphics (Siggraph '85), vol. 19, No. 3, pp. 31-40, 1985.
G.A. Crocker, "Invisibility coherence for faster scan-line hidden surface algorithms", Computer Graphics, vol. 18, No. 3, pp. 95-102, 1984.
H. Hubschman et al., "Frame-to-frame coherence and the hidden surface computations: Constraints for a Convex World", Computer Graphics, vol. 15, No. 3, pp. 45-54, 1981.
H.L. Lim, "Fast hidden surface removal through structural analysis and representation of objects and their contours", Computer Graphics International'87, pp. 75-88, 1987.
H.L. Lim, "Toward a fuzzy hidden surface algorithm", Computer Graphics International '92, pp. 621-635, 1992.
H.L. Lim, "An efficient hidden surface algorithm for polyhedral surfaces" in International Conference on Computer & Communications in Science & Technology, Beijing, China, 1986.
C. Hornung, "A method for solving the visibility problem", IEEE Computer Graphics & Applications, vol. 4, pp. 26-33, 1984.
J. Griffiths, "A depth-coherence scanline algorithm for displaying curved surfaces", Computer-aided Design, vol. 16, No. 2, pp. 91-101, 1984.
E.A. Haines et al., "Shaft culling for efficient ray-traced radiosity", Proceedings of Eurographics Workshop on Rendering, Jul., 1991.
H. Plantinga et al., "Real-time hidden-line elimination for a rotating polyhedral scene using the aspect representation", in Graphics Interface '90 Proceedings, pp. 9-16, 1990.
J.M. Airey et al., "Towards image realism with interactive update rates in complex virtual building environment", in ACM Siggraph Special Issue on the 1990 Symposium on Interactive 3D Graphics vol. 24, pp. 41-50, 1990.
Y. Wang, "Image synthesis using front-to-back based radiosity methods", Ph.D. thesis, University of Alberta, 1992.
D.P. Luebke et al., "Portal and mirrors: simple, fast evaluation of potentially visible sets", in Proceedings 1995 Symposium on Interactive 3-D Graphics, ACM Press, New York, 1995.
H. Plantinga, "Conservative visibility preprocessing for efficient walkthroughs of 3D scenes", Graphics Interface '93 Proceedings, pp. 166-173, 1993.
E. Catmull, "A subdivision algorithm for computer display of curved surfaces", Ph.D. thesis, Utah University, Dec. 1974.
J. Arvo et al., "A survey of ray tracing acceleration techniques" in An Introduction to Ray Tracing, editor: A.S. Glassner, Academic Press, London, pp. 201-262, 1989.
A. Watt et al., "Advanced Animation and Rendering Techniques, Theory and Practice", ACM Press, New York, 1992.
F. Sillion et al., "Radiosity and Global Illumination", Morgan Kaufmann Publisher, San Francisco, 1994.
D.R. Baum et al., "Improving radiosity solutions through the use of analytically determined form-factors", Computer Graphics, vol. 23, No. 3, pp. 325-334, 1989.
A. Fournier et al., "On the power of the frame buffer", ACM Transaction on Graphics, vol. 7, No. 2, 103-128, 1988.
C.W. Grant, "Integrated analytic spatial & temporal anti-aliasing for polyhedra in 4-Space", Computer Graphics, vol. 19, No. 3, pp. 79-84, 1985.
P. Hsiung et al., "T-Buffer: fast visualization of relativistic effects in spacetime", ACM Siggraph Special Issue on the 1990 Symposium on Interactive 3D Graphics 24, pp. 83-88, 1990.
A. Inselberg, "The Plane with parallel coordinates", The Visual Computer, vol. 1, pp. 69-91, 1985.
N.L. Max et al., "A two-and-a-half-D motion-blur algorithm", Computer Graphics (Siggraph '85), vol. 19, No. 3, pp. 85-93, 1985.
K.V. Steiner et al., "Hidden volumes: the 4th dimension", Computer Graphics World, pp. 71-74, Feb. 1987.
L.A. Zadeh, "Fuzzy Sets", Information and Control, vol. 8, pp. 338-353, 1965.
W. Siedleckiet al., "Mapping techniques for exploratory pattern analysis" in Pattern Recognition and Artificial Intelligence, E.S. Gelsema, L.N. Kanal (eds), Elsevier, New York, pp. 277-299, 1988.
S. Chang, "On fuzzy mapping and control", IEEE Transactions on Systems, Man & Cybernetics, vol. SMC-2, No. 1, pp. 30-34, 1972.
R. Jain et al., "Imprecision in computer vision" in Advances in Fuzzy Sets, Possibility and Applications, P. Wang, (ed), Plenum Press, New York, pp. 217-236, 1983.
N. Greene et al., "Hierarchical z-buffer visibility", Computer Graphics (Siggraph '93), vol. 27, pp. 231-238, 1993.
D. Greenberg et al., "Radiosity: a method for computing global illumination", The Visual Computer, vol. 2, pp. 291-297, 1986.
H. Fuchs et al., "New real-time shaded display of rigid objects", Computer Graphics, vol. 17, No. 3, pp. 65-72, 1983.
C.M. Hoffman et al., "Some techniques for visualizing surfaces in four-dimensional space", Computer-aided Design, vol. 23, No. 1, pp. 83-91, 1991.
J.M. Lane et al., "Scan line methods for displaying parametrically defined surfaces", Communications of the ACM, vol. 23, No. 1, pp. 23-34, 1980.
J.A. Gualtieri et al., "The visual potential: one convex polygon", Computer Vision, Graphics and Image Processing, vol. 46, No. 1, pp. 96-130, 1989.
Z. Gigus et al., "Computing the aspect graph for line drawings of polyhedral objects", in Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Computer Society Press, New York, pp. 654-661, 1988.
P.S. Heckbert et al., "Beam tracing polygonal objects", Computers Graphics, vol. 18, No. 3, pp. 119-127, 1984.
J. Zhou, "Visualization of four dimensional space and its applications", Ph.D. thesis, Purdue University, 1991.
R. Jain, "Application of fuzzy sets for the analysis of complex scenes" in Advances in Fuzzy Set Theory and Applications, M.M. Gupta et al. (eds), North-Holland, Amsterdam, pp. 577-583, 1979.
D. Tost et al., "A definition of frame-to-frame coherence", Computer Animation '90, pp. 207-221, 1990.
Y. Leung, "Spatial Analysis and Planning under Imprecision", North Holland, Amsterdam, 1988.
J.R. Wallace et al, "A ray tracing algorithm for progressive radiosity", Computer Graphics, vol. 23, No. 3, pp. 315-324, 1989.
H. Fuchs et al., "On visible surface generation by a priori tree structure", Computer Graphics (Siggraph'80), vol. 14, pp. 124-133, 1980.
M.F. Cohn et al., "A progressive refinement approach to fast radiosity image generation", Computer Graphics, vol. 22, No. 4, pp. 75-84, 1988.
K. Kanatani, "Group-theoretical Methods in Image Understanding", Springer Verlag, Berlin, 1990.
J.K. Aggarwal et al., "Dynamic scence analysis" in Image Sequence Processing and Dynamic Scene Analysis, T.S. Huang (ed), Springer Verlag, Berlin, pp. 40-73, 1983.
N.I. Badler et al., "Motion: Representation and Perception" Elsevier, New York, 1986.
Subbarao, "Interpretation of visual motion: A computational study", Pitman, London, 1988.
F. Sillion et al., "A general two-pass method integrating specular and diffuse reflection", Computer Graphics, vol. 23, No. 3, pp. 335-344, 1989.
R.J. Recker et al., "Acceleration Techniques for Progressive Refinement Radiosity", ACM Siggraph Special Issue on the 1990 Symposium on Interactive 3D Graphics, vol. 24, pp. 59-66, 1990.
E. H. Ruspini, "A new approach to clustering", Information and Control 15, pp. 22-32, 1969.
A. Kaufmann, "Theory of Fuzzy Subsets. vol. I, Fundamental Theorectical Elements", Academic Press, London, 1975.
T.S. Huang, "Image Sequence Processing", Springer Verlag, Berlin, 1981.
P.A. Ligomenides, "Modeling uncertainty in human perception" in Uncertainty in Knowledge-Based Systems, B. Bouchon, R. Yager (eds), Springer Verlag, Berlin, pp. 337-346, 1986.
A. Inselberg, "N-Dimensional Graphics. Part I. Lines & Hyperplanes", IBM Scientific Center Report G320-2711, Jul. 1981.
A.S. Glassner, "3D Computer Graphics: A User's Guide for Artists & Designers", 2nd edition, Design Press, New York, pp. 139-158, 1989.
M.F. Cohen et al., "Radiosity & realistic image synthesis", Academic, New York, 1993.
J. Aggarwal et al., "Analysing dynamic scenes containing multiple moving objects" in Image Sequence Analysis, editor: T.S. Huang, Springer-Verlag, Berlin, pp. 355-380, 1981.
A. Appel, "The Notion of quantitative invisibility and the machine rendering of solids", Proceedings ACM National Conference, Thompson Books, Washington, DC, pp. 387-393, 1967.
J. Vince, "Computer Animation", Addison-Wesley, New York, 1992.
Y. Chrysanthou et al., "Computing dynamic changes to BSP-trees", Eurographics '92, vol. 11, No. 3, pp. C-321-C-332, 1992.
S. Ansoldi et al., "Geometric modeling of solid objects by using a face adjacency graph presentation" Computer Graphics (Siggraph '85), vol. 19, No. 3, pp. 131-138, 1985.
N. Chin et al., "Fast object-precision shadow generation for area light sources using BSP trees", Proceedings 1992 Symposium on Interactive 3D Graphics, pp. 21-30, 1992.
D.S. Immel et al., "A radiosity method for non-diffuse environments", Comuter Graphics, vol. 4, pp. 133-142, 1986.
P. Schuytema How to bake a quake Computer Gaming Jul. 1996.
Efficient Collision Detection for Animation and Robotics Ming C. Lin 1993.
Fast Computation of Shadow Boundaries Using Spatial Coherence and Backprojections A James Stewart Sherif Ghali.
Polygon shadow generation Atherton P. Weiller K. Greenberg D. Proceedings, Siggraph'78.
Designing real-time 3d graphics for entertainment Siggraph'96 course notes.
Foley, "Computer Graphics: Principles and Practice, Second Edition" Addison-Wesley Publishing Company, Inc. pp. 663-665, 772, 783, 793-797, 805 (1990).
Sutherland, "A Characterization of Ten Hidden-Surface Algorithms," Computing Surveys, vol. 6, No. 1, pp. 38-39, Mar. 74.

Visibility calculations for 3D computer graphics

Information

Patent Number

Date Filed

Date Issued

Inventors

Examiners

Agents

CPC

US Classifications

Field of Search

US

International Classifications