Patent Application
20070291143
The invention relates to a system for the production of images, the system comprising one or more cameras for providing overlapping images on a base, the system further comprising a merger for merging overlapping images supplied by the one or more of the cameras into a composite image.
The invention also relates to a method for the production of images using one or more cameras wherein overlapping images are produced and overlapping images are merged into a composite image.
Such a system and method is known from U.S. Pat. No. 5,657,703. In said patent a system and method is described. The described image system comprises a plurality of video cameras on a base, each of which cameras captures an image, the data of each image forming a stream of digital or analog output. The images overlap and the streams are merged to form a panoramic or panospheric image. The system and device also comprises an image clipper to select a portion of the overall image. This portion may then be displayed on a display. Advances in digital storage and wireless communication make it possible to use such a video stream for entertainment.
This known system uses a fixed base.
Cameras, especially digital cameras, are becoming smaller, but also cheaper and more robust. This makes it possible to place them in almost any objects. The same type of solutions is breaking also into the consumer market.
In modern TV entertainment it becomes more and more interesting to place cameras in unusual locations which provide unique views such as e.g. bicycles during a race, formula I driver helmets, guitars. Such objects move.
The inventors have realized that especially for sports, but not exclusively for sport, rotatable objects are very interesting objects to place a camera in or upon, i.e. to be used as a base for the cameras. Examples are for instance balls such as soccer, American Football, basketball balls. Placing a camera on such rotatable objects, i.e. on a rotatable base provides an unique perspective, enabling a view direct from the centre of activity rather than from an outside point of view. Thus a system in which base object for the one or more cameras is a rotatable object, i.e. an object which when used is liable to rotate.
For such a system number of requirements should preferably be met:
The system as described in U.S. Pat. No. 5,657,703 is directed to the use of non-moving cameras. The base is in essence a fixed camera pedestal and the camera have a fixed point of view. For cameras mounted on rotatable objects, such as balls, the cameras move with respect to the image taken, even if they do not move with respect to the base (e.g. the ball) to which they are attached. The field of view of each camera changes causing apparent movement of objects within the field of view. Such apparent movement, even for relatively slow rotational speed may be large. This apparent movement may cause problems in merging (or stitching) of the overlapping images taken by the camera or cameras into one large image. “Overlapping” within the framework of the invention means that at least some parts of images comprise the same content matter. A priori the rotation (rotational speed, rotational axis) as well as the translational speed and direction of the object is unknown which complicates matters.
The present invention aims to provide a system and a method in or for which at least some of the mentioned problems are reduced.
The invention in its various aspects is based on the following insight.
When a static arrangement of cameras as in U.S. Pat. No. 5,657,073 is used the field of views and the overlap of the field of views covered by various cameras are relatively well defined. Thus the overlap of images is relatively well defined. The same, however, is not true when the camera or cameras are positioned on a rotating object such as a ball. The field of view of the camera(s) changes constantly in an a priori unknown manner. Furthermore cameras do not take pictures continuously but intermittently, for instance 50 images per second. Even at relatively small or moderate rotational speeds, for instance 2 rotations per second, a number of problems occur, which do not occur with a stationary plurality of cameras. One of these problems is the occurrence of blind spots in the panoramic image, i.e. parts of the overall image where inadvertently no image is taken during at least some time period. Another is an uncertainty of the field of overlap. The fields of overlap will constantly vary in an a priori unknown fashion.
To reduce such problems the system in accordance with a first aspect of the invention is characterized in that it comprises an image frequency determinator for determining the frequency at which images are taken by the one or more cameras in dependence on the rotation of the base.
A priori the rotation of the base is unknown. The larger the rotational speed, i.e. the number of rotations per second of the base to which the camera or cameras are attached, the larger the problems. By determining the frequency with which images are taken as a function of the rotation, more in particular in general increasing the frequency with which images are taken, i.e. the number of images taken per second, increases as the rotation increase. Taking more images as the object rotates faster increases the correspondence between subsequent images taken by a camera, making merging of images easier.
The frequency with which images are taken may be linearly dependent on the rotation of the object, or the camera may have a number of settings, e.g. low, middle and high speed. In embodiments wherein the system comprises a plurality of cameras the determinator may determine a common image frequency for all of the cameras, or in more sophisticated embodiments of the system the determinator may comprise a discriminator for providing differing image frequencies for different cameras.
The rotation of the object may be determined from the images taken, for instance the rotational frequency may be determined by the frequency with which a particular object is present in images taken by one of the camera, or it may be determined by a sensor within the rotatable object, for instance a strain gauge measuring centrifugal forces within the object as it rotates, or it may be determined by a sensor positioned outside the object, for instance a fixed separate camera following the object, wherein the object is provided with a marking, and the sensor measured the rotational speed by following the movement of this marking.
In simple embodiments the image frequency determinator is arranged for determining the frequency at which images are taken by the one or more cameras in dependence on the rotation frequency of the base. In more sophisticated embodiments the determinator is arranged for determining the frequency at which images are taken by the one or more cameras in dependence on the rotational frequency of the base and in addition on the orientation of the rotational axis and/or the translational speed.
In an embodiment the system comprise a selector for selection the resolution of the image as a function of image frequency. As the image frequency increases the total amount of data increases. By reducing the accuracy of the image, i.e. taking images at a lower resolution the total amount of data can still be kept in bounds. Doubling the frequency when the rotation is doubled while halving the resolution will still give an improved results, since, although the total amount of data per second stays the same, the difference between images is less and merging is easier.
This aspect of the invention is related to the frequency with which images are taken, i.e. timing of the time delay between images taken by a same camera as a function of rotational speed. This resolves at least some of the problems encountered in merging of images. This aspect of the invention relates to systems having one camera as well as to systems have a plurality of cameras. Based upon the same insight the inventors have realized that in systems where a plurality of cameras are used, advantages may be obtained by timing the taking of images taken by different cameras.
The system in accordance a second aspect of the invention is characterized in that system comprises a plurality of cameras and the system comprises a timer for mutually timing the instances at which images are taken at respective cameras.
‘mutually’ within the framework of the invention means that the instances at which images are taken are functionally related to each other and predetermined before the images are taken.
By using a timer to time the taking of images of different cameras, blind spots and uncertainty in the extent of the overlap of the fields of view can be avoided or at least reduced. Blind spots and varying overlaps of fields of view occur when a first and second adjacent camera whose instantaneous fields of view overlap take images some time apart. During this time the object may have rotated such that in fact there is no longer an overlap between the taken images, causing a blind spot or a much larger or varying overlap. A temporary blind spot in the overall image has then occurred. Such a blind spot makes merging of the image difficult. Varying fields of overlap also cause problems. By determining the times at which the image are taken and doing so coherently the occurrence of blind spots and problems with varying extent of overlap of fields of view may be avoided or at least reduced.
In a simple arrangement the timer is arranged such that in operation all cameras of the plurality of cameras take images simultaneously. In this simple arrangement the timer is arranged for synchronising at least as set of cameras, such that for said set taking of the images is synchronised. This reduces the occurrence of blind spots. The object cannot rotate between the taking of images by different cameras of the set of cameras, because the images are taken simultaneously.
In a variation on this arrangement the set of cameras comprises a number of subsets, wherein the timer is arranged such that in operation for each subset the cameras taken image simultaneously, but the timing of the subsets differ.
An example of such an arrangement is a ball having three subsets of 6 cameras, each subset providing a panospheric image. Suppose images are taken 50 times per second. If all subsets take images simultaneously 50 overall image of high quality is taken. If the timing of the subsets differ (for instance by 1/150 of a second) 150 overall images are taken (be it with fewer cameras). For a fast moving or rotating object the latter arrangement provides better results. Coherent timing of the instances at which the images for the various cameras are taken provides such advantage.
The system in preferred embodiments comprises a means for determining a rotational speed and the timer is arranged to vary timing in dependence of the determined rotational speed.
Variation of timing may be for instance such that the frequency of taking of images is dependent on the rotational speed, i.e. the higher the rotational speed, the more images are taken per second, as in the first aspect of the invention. The higher the rotational speed the higher the apparent speed of objects in the image. Increasing the frequency with which images are taken will increase image quality.
Variation of timing may also be done in a slightly different fashion. Merging of the objects can best be done by finding common objects within separate images. For stationary cameras such common objects are relatively easily found. However, for rotation objects, where the field of view changes drastically, such may not be so easy. By determining the rotational speed and comparing it to the timing, a course comparison of the field of views may be made to make a coarse determination of where in various images common objects may be found, which helps in merging the images into an overall image. This embodiment is in particular of importance when the system comprises a means, such as a clipper, to select a portion of the overall image. In contrast with the known system where the portion is stationary or quasi-stationary with respect to the fields of views of the cameras, such is not the case when the cameras are mounted on a rotating object such as a ball. For a pedestal viewing a playing field the position in respect of the various cameras of a goalkeeper will be known to some degree. However, for a rotating ball speeding towards a goalkeeper it will a priori not be known which camera captures the goalkeeper. By determining on the one hand the rotational speed (and possible also the overall speed) and on the other hand the instantaneous position of the goalkeeper within a field of view, the system in accordance with this embodiment times the taking of pictures for instance such that the goalkeeper is in the centre of fields of view when images are taken.
Rotational speed, as described above, may be determined by a rotational speed determinator, such as for instance a device which by measuring centrifugal forces determines the rotational axis or axes and the speed of rotation, or the rotational speed may be determined from previously taken images, wherein comparison of a sequence of images by finding common stationary objects allows the speed of rotation to be determined.
In embodiments the time differences, the rotational speed and or the translational speed, form inputs for the merger.
In embodiments the system, whether in the first or second aspect, may comprise sensors for determining deformation of the base.
A ball, when hit, may deform. This deformation will have its effects on the fields of view of the various cameras. By measuring deformation, for instance by sensors, information on the deformation is obtained, which information forms an input for the merger to be used in the merging of the image, or in the timing of taken images.
In an embodiment of the invention the system is arranged to generate a field of view independent of the actual images, when deformation exceeds a threshold.
When a ball is hit, the deformation may be so large that for a brief instance the image is either difficult to form, or in fact a true-to-life image is very confusing.
In preferred embodiments the system is arranged to provide a smooth transition between the image just prior to the hitting of the ball and an image a short time after hitting the ball.
The system in accordance with a second embodiment of second aspect of the invention is characterized in that it comprises a timer for establishing the timing the instances at which images have been taken at respective cameras.
“Establishing” within the framework of the invention means that after the images are taken, the time difference between the taking of images between various cameras is established. To some extent this embodiment is complementary to the first embodiment. However, instead of predetermining the timing, the system determines after the fact what time differences have occurred between various cameras. These time differences are an input to the merger. The invention encompasses those preferred embodiments described in respect of the first aspect in combination with the second aspect, in so far as such measures are commensurable with the second aspect. Sensors for deformation, sensors or other means for determination of rotational speed, means for selecting of a the overall image at least may be used for this second embodiment of the second aspect.
The first and second aspect relate in the various embodiments all relate to timing of the taking of images. The third aspect of the invention relates to the merger itself.
The system in accordance with a third aspect of the invention is characterized in that the merger comprises a motion estimator for estimating motion vectors between images, wherein the motion estimator is so arranged that it has an input for a rotational parameter of the base.
Motion estimation is a known technique in which images are compared to find common objects and motion vectors describing the apparent motion of such an object between the compared images.
For such a motion estimation usually use is made of a library of motion vectors. The inventors have realized that known motion estimation techniques would require a very large range in possible motion vectors. In fixed systems the motion of object is rather limited. Most of the image is static and only a few objects move within the image. Even when motion estimation is used for for instance compensating a shaking of the image, or panning of an image the motion is rather simple, the whole image is shaken or moves in a more or less regular manner. So the choice of motion vectors is rather limited and usually relatively simple. However, the inventors have realized that when the base is rotating at an a priori unknown rotational frequency and a priori unknown translational speed the situation is very different. The apparent motion of objects varies widely in time and space. Consequently the motion estimation is much more difficult. In fact this also relates to the timing of images. The differences between images to be merged varies widely. It is not a priori known how much motion there is due to the unknown rotation of the object. In the third aspect of the invention the characteristics of the rotation (rotational speed and/or orientation of rotational axis) form an input for the motion estimator. The rotational parameters make it possible to direct a range finder to an set of possible motion vectors thus requiring less time and effort to estimate the motion. Beside the characteristics of the rotation of the base, the motion estimator may also have an input for characteristics of the translation (speed and/or direction). In effect this third aspect of the invention also relates to timing, the input of the rotational speed combined with knowledge of time delays (i.e. timing) between images (whether between consecutive images taken by one and the same camera or images taken by different cameras) will provided information allowing first order estimation of motion vectors. This allows the motion estimator to determine, starting from this first order approximation for the motion vector relatively fast the actual motion vectors.
In a preferred embodiment the estimation techniques are independent of the orientation of the camera. In standard motion estimation techniques the estimator uses different algorithms for the horizontal direction and the vertical direction. Basically it is usually assumed that there is a much larger probability for a motion in a horizontal direction then for a vertical direction. Cameras are always held and oriented as they are supposed to be held (i.e. bottom part down) and most movements are from left to right and vice versa. Thus is make sense for the known motion estimation systems to use the available computing power in the most efficient way and discriminate between horizontal and vertical motion.
However, for a rotatable object, it is not known how the cameras will be oriented, and all directions should preferably be treated on an equal basis.
Preferably the cameras take images with equal resolution and extent in horizontal and vertical direction.
These and further aspects of the invention will be explained in greater detail by way of example and with reference to the accompanying drawings, in which
The Figs. are not drawn to scale. Generally, identical components are denoted by the same reference numerals in the figs.
The system comprises a timer 7 to time the taking of images of the cameras 2, 34 etc. The timer in the first aspect of the invention is a determinator for determining the frequency (f) of the images, i.e. how many images are taken per second and/or a timer to determine the time (Δt) between images taken. To this end the timer has an input for the rotational frequency ω or other data relating to said rotation.
The rotation of the object may be determined by the images taken, for instance the rotational frequency c may be determined by the frequency with which a particular object is present in images taken by one of the camera, it may also be determined by a sensor within the rotatable object, for instance a strain gauge within a ball measuring centrifugal forces within the ball as it rotates, or it may be determined by a sensor positioned outside the object, for instance a fixed separate camera following the object, wherein the object is provided with a marking, and the sensor measured the rotational speed by following the movement of this marking.
In simple embodiments the image frequency determinator (7) is arranged for determining the frequency at which images are taken by the one or more cameras in dependence on the rotational frequency of the base. In more sophisticated embodiments the determinator is arranged for determining the frequency at which images are taken by the one or more cameras in dependence on the rotational frequency of the base and/or the orientation of the rotational axis and/or the translational speed wherein the determinator increases the image frequency the higher the translational speed. Translational speed may be determined from the images taken previously, a sensor within the base (for instance a sensor as used in airplanes to measure ground speed) or by means of a separate camera or speed gun.
The system and method in accordance with a second aspect of the invention relates to mutual timing of the taking of images by different cameras.
In a simple embodiment a simple timing arrangement is used, which is independent of the rotational frequency of the base: images are taken simultaneously. By synchronizing the images of the various cameras at least one disadvantageous effect of rotation is removed, since there is no time difference Δt between the images and the extent of overlap of images taken by adjacent cameras is at least in first order approximation known. This reduces problems with merging of the images, and avoids the occurrence of black spots.
A preferred embodiment is schematically shown in
The system comprises two sub-sets of cameras 2, 3, 4 and 2a, 3a, 4a. Within both sub-sets the cameras are synchronized. However, there is a time difference between the synchronization of both subsets. This allows the system to provide a higher rate of taking of images. The time difference between the sets may be dependent on the speed of rotation wherein, which in effect would constitute a system and method in which first and second aspect of the invention are combined. For instance the time difference between the subsets is zero at low rotational speed (allowing to take very high definition images) but at higher rotational speeds there is an interval between the triggering of both subsets equal to half the interval between images within a subset which allows for a better image reproduction.
In
The device may comprise a device for measuring deformation. The device for measuring rotational speed may double as a device for measuring deformation for instance when strains are measured. Strains indicated force on the ball, such forces may indicate deformations as well as centrifugal forces. Because deformations cause cameras to be moved in respect of each other, information on deformations is useful information for merging of images.
The above aspect of the invention relate to timing of the taking of overlapping images before merging said image into a composite image is performed.
A third aspect of the invention relates to the merger itself. In this aspect the merger comprises a motion estimator, i.e. an algorithm with which apparent motion of common objects in different images is estimated. Such estimation helps in merging of the images into the composite image. The inventors have realized that for cameras on rotatable objects data on the timing of the images, in combination with rotational speed, is important and can be used to advantage.
Motion estimation is a known technique in which images are compared to find common objects and motion vectors describing the apparent motion of such an object between the compared images.
For such a motion estimation usually use is made of a library of motion vectors. The inventors have realized that known motion estimation techniques would require a very large range in possible motion vectors. In fixed systems the motion of object is rather limited. Most of the image is static and only a few objects move within the image. Even when motion estimation is used for for instance compensating a shaking of the image, or panning of an image the motion is rather simple, the whole image is shaken or moves in a more or less regular manner. So the choice of motion vectors is rather limited and usually relatively simple. However, the inventors have realized that when the base is rotating at an a priori unknown rotational frequency and a priori unknown translational speed the situation is very different. The apparent motion of objects varies widely in time and space. Consequently the motion estimation is much more difficult. In fact this also relates to the timing of images. The differences between images to be merged varies widely. It is not a priori known how much motion there is due to the unknown rotation of the object. In the third aspect of the invention the characteristics of the rotation (rotational speed and/or orientation of rotational axis) form an input for the motion estimator. The rotational parameters make it possible to direct a range finder to a set of possible motion vectors thus requiring less time and effort to estimate the motion. Beside the characteristics of the rotation of the base, the motion estimator may also have an input for characteristics of the translation (speed and/or direction). In effect this third aspect of the invention also relates to timing, the input of the rotational speed combined with knowledge of time delays (i.e. timing) between images (whether between consecutive images taken by one and the same camera or images taken by different cameras) will provide information allowing first order estimation of motion vectors. This allows the motion estimator to determine, starting from this first order approximation for the motion vector relatively fast the actual motion vectors.
Within the concept of the invention a ‘merger” and “timer” as well as “means for recording” etc. etc. is to be broadly understood and to comprise e.g. any piece of hard-ware (such a merger, timer, recorder), any circuit or sub-circuit designed for merging images, time the taking of images, recording etc. as described as well as any piece of soft-ware (computer program or sub program or set of computer programs, or program code(s)) designed or programmed to take such action in accordance with the invention as well as any combination of pieces of hardware and software acting as such, alone or in combination, without being restricted to the above or below given exemplary embodiments.
The invention is also embodied in any computer program comprising program code means for performing a method in accordance with the invention when said program is run on a computer as well as in any computer program product comprising program code means stored on a computer readable medium for performing a method in accordance with the invention when said program is run on a computer, as well as any program product comprising program code means for use in a system in accordance with the invention, for performing the action specific for the invention.
It will be appreciated by persons skilled in the art that the present invention is not limited by what has been particularly shown and described hereinabove. The invention resides in each and every novel characteristic feature and each and every combination of characteristic features. Reference numerals in the claims do not limit their protective scope. Use of the verb “to comprise” and its conjugations does not exclude the presence of elements other than those stated in the claims. Use of the article “a” or “an” preceding an element does not exclude the presence of a plurality of such elements.
| Number | Date | Country | Kind |
|---|---|---|---|
| 04104663.2 | Sep 2004 | EP | regional |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/IB05/53035 | 9/15/2005 | WO | 3/19/2007 |
