Patent Application
20230154052
Point cloud content is content represented by a point cloud, which is a set of points belonging to a coordinate system representing a three-dimensional space. The point cloud content may express media configured in three dimensions, and is used to provide various services such as virtual reality (VR), augmented reality (AR), mixed reality (MR), and self-driving services. However, tens of thousands to hundreds of thousands of point data are required to represent point cloud content. Therefore, there is a need for a method for efficiently processing a large amount of point data.
Embodiments provide a device and method for efficiently processing point cloud data. Embodiments provide a point cloud data processing method and device for addressing latency and encoding/decoding complexity.
The technical scope of the embodiments is not limited to the aforementioned technical objects, and may be extended to other technical objects that may be inferred by those skilled in the art based on the entire contents disclosed herein.
To achieve these objects and other advantages and in accordance with the purpose of the disclosure, as embodied and broadly described herein, a method for transmitting point cloud data may include encoding the point cloud data, and transmitting a bitstream containing the point cloud data.
Devices and methods according to embodiments may process point cloud data with high efficiency.
The devices and methods according to the embodiments may provide a high-quality point cloud service.
The devices and methods according to the embodiments may provide point cloud content for providing general-purpose services such as a VR service and a self-driving service.
The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the disclosure and together with the description serve to explain the principle of the disclosure. For a better understanding of various embodiments described below, reference should be made to the description of the following embodiments in connection with the accompanying drawings. The same reference numbers will be used throughout the drawings to refer to the same or like parts. In the drawings:
Reference will now be made in detail to the preferred embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. The detailed description, which will be given below with reference to the accompanying drawings, is intended to explain exemplary embodiments of the present disclosure, rather than to show the only embodiments that may be implemented according to the present disclosure. The following detailed description includes specific details in order to provide a thorough understanding of the present disclosure. However, it will be apparent to those skilled in the art that the present disclosure may be practiced without such specific details.
Although most terms used in the present disclosure have been selected from general ones widely used in the art, some terms have been arbitrarily selected by the applicant and their meanings are explained in detail in the following description as needed. Thus, the present disclosure should be understood based upon the intended meanings of the terms rather than their simple names or meanings.
The point cloud content providing system illustrated in
The point cloud data transmission device 10000 according to the embodiments may secure and process point cloud video (or point cloud content) and transmit the same. According to embodiments, the transmission device 10000 may include a fixed station, a base transceiver system (BTS), a network, an artificial intelligence (AI) device and/or system, a robot, an AR/VR/XR device and/or server. According to embodiments, the transmission device 10000 may include a device, a robot, a vehicle, an AR/VR/XR device, a portable device, a home appliance, an Internet of Thing (IoT) device, and an AI device/server which are configured to perform communication with a base station and/or other wireless devices using a radio access technology (e.g., 5G New RAT (NR), Long Term Evolution (LTE)).
The transmission device 10000 according to the embodiments includes a point cloud video acquirer 10001, a point cloud video encoder 10002, and/or a transmitter (or communication module) 10003.
The point cloud video acquirer 10001 according to the embodiments acquires a point cloud video through a processing process such as capture, synthesis, or generation. The point cloud video is point cloud content represented by a point cloud, which is a set of points positioned in a 3D space, and may be referred to as point cloud video data. The point cloud video according to the embodiments may include one or more frames. One frame represents a still image/picture. Therefore, the point cloud video may include a point cloud image/frame/picture, and may be referred to as a point cloud image, frame, or picture.
The point cloud video encoder 10002 according to the embodiments encodes the acquired point cloud video data. The point cloud video encoder 10002 may encode the point cloud video data based on point cloud compression coding. The point cloud compression coding according to the embodiments may include geometry-based point cloud compression (G-PCC) coding and/or video-based point cloud compression (V-PCC) coding or next-generation coding. The point cloud compression coding according to the embodiments is not limited to the above-described embodiment. The point cloud video encoder 10002 may output a bitstream containing the encoded point cloud video data. The bitstream may contain not only the encoded point cloud video data, but also signaling information related to encoding of the point cloud video data.
The transmitter 10003 according to the embodiments transmits the bitstream containing the encoded point cloud video data. The bitstream according to the embodiments is encapsulated in a file or segment (for example, a streaming segment), and is transmitted over various networks such as a broadcasting network and/or a broadband network. Although not shown in the figure, the transmission device 10000 may include an encapsulator (or an encapsulation module) configured to perform an encapsulation operation. According to embodiments, the encapsulator may be included in the transmitter 10003. According to embodiments, the file or segment may be transmitted to the reception device 10004 over a network, or stored in a digital storage medium (e.g., USB, SD, CD, DVD, Blu-ray, HDD, SSD, etc.). The transmitter 10003 according to the embodiments is capable of wired/wireless communication with the reception device 10004 (or the receiver 10005) over a network of 4G, 5G, 6G, etc. In addition, the transmitter may perform a necessary data processing operation according to the network system (e.g., a 4G, 5G or 6G communication network system). The transmission device 10000 may transmit the encapsulated data in an on-demand manner
The reception device 10004 according to the embodiments includes a receiver 10005, a point cloud video decoder 10006, and/or a renderer 10007. According to embodiments, the reception device 10004 may include a device, a robot, a vehicle, an AR/VR/XR device, a portable device, a home appliance, an Internet of Things (IoT) device, and an AI device/server which are configured to perform communication with a base station and/or other wireless devices using a radio access technology (e.g., 5G New RAT (NR), Long Term Evolution (LTE)).
The receiver 10005 according to the embodiments receives the bitstream containing the point cloud video data or the file/segment in which the bitstream is encapsulated from the network or storage medium. The receiver 10005 may perform necessary data processing according to the network system (for example, a communication network system of 4G, 5G, 6G, etc.). The receiver 10005 according to the embodiments may decapsulate the received file/segment and output a bitstream. According to embodiments, the receiver 10005 may include a decapsulator (or a decapsulation module) configured to perform a decapsulation operation. The decapsulator may be implemented as an element (or component) separate from the receiver 10005.
The point cloud video decoder 10006 decodes the bitstream containing the point cloud video data. The point cloud video decoder 10006 may decode the point cloud video data according to the method by which the point cloud video data is encoded (for example, in a reverse process of the operation of the point cloud video encoder 10002). Accordingly, the point cloud video decoder 10006 may decode the point cloud video data by performing point cloud decompression coding, which is the inverse process of the point cloud compression. The point cloud decompression coding includes G-PCC coding.
The renderer 10007 renders the decoded point cloud video data. The renderer 10007 may output point cloud content by rendering not only the point cloud video data but also audio data. According to embodiments, the renderer 10007 may include a display configured to display the point cloud content. According to embodiments, the display may be implemented as a separate device or component rather than being included in the renderer 10007.
The arrows indicated by dotted lines in the drawing represent a transmission path of feedback information acquired by the reception device 10004. The feedback information is information for reflecting interactivity with a user who consumes the point cloud content, and includes information about the user (e.g., head orientation information, viewport information, and the like). In particular, when the point cloud content is content for a service (e.g., self-driving service, etc.) that requires interaction with the user, the feedback information may be provided to the content transmitting side (e.g., the transmission device 10000) and/or the service provider. According to embodiments, the feedback information may be used in the reception device 10004 as well as the transmission device 10000, or may not be provided.
The head orientation information according to embodiments is information about the user' s head position, orientation, angle, motion, and the like. The reception device 10004 according to the embodiments may calculate the viewport information based on the head orientation information. The viewport information may be information about a region of a point cloud video that the user is viewing. A viewpoint is a point through which the user is viewing the point cloud video, and may refer to a center point of the viewport region. That is, the viewport is a region centered on the viewpoint, and the size and shape of the region may be determined by a field of view (FOV). Accordingly, the reception device 10004 may extract the viewport information based on a vertical or horizontal FOV supported by the device in addition to the head orientation information. Also, the reception device 10004 performs gaze analysis or the like to check the way the user consumes a point cloud, a region that the user gazes at in the point cloud video, a gaze time, and the like. According to embodiments, the reception device 10004 may transmit feedback information including the result of the gaze analysis to the transmission device 10000. The feedback information according to the embodiments may be acquired in the rendering and/or display process. The feedback information according to the embodiments may be secured by one or more sensors included in the reception device 10004. According to embodiments, the feedback information may be secured by the renderer 10007 or a separate external element (or device, component, or the like). The dotted lines in
According to embodiments, the transmission device 10000 may be called an encoder, a transmission device, a transmitter, or the like, and the reception device 10004 may be called a decoder, a receiving device, a receiver, or the like.
The point cloud data processed in the point cloud content providing system of
The elements of the point cloud content providing system illustrated in
The block diagram of
The point cloud content providing system according to the embodiments (for example, the point cloud transmission device 10000 or the point cloud video acquirer 10001) may acquire a point cloud video (20000). The point cloud video is represented by a point cloud belonging to a coordinate system for expressing a 3D space. The point cloud video according to the embodiments may include a Ply (Polygon File format or the Stanford Triangle format) file. When the point cloud video has one or more frames, the acquired point cloud video may include one or more Ply files. The Ply files contain point cloud data, such as point geometry and/or attributes. The geometry includes positions of points. The position of each point may be represented by parameters (for example, values of the X, Y, and Z axes) representing a three-dimensional coordinate system (e.g., a coordinate system composed of X, Y and Z axes). The attributes include attributes of points (e.g., information about texture, color (in YCbCr or RGB), reflectance r, transparency, etc. of each point). A point has one or more attributes. For example, a point may have an attribute that is a color, or two attributes that are color and reflectance. According to embodiments, the geometry may be called positions, geometry information, geometry data, or the like, and the attribute may be called attributes, attribute information, attribute data, or the like. The point cloud content providing system (for example, the point cloud transmission device 10000 or the point cloud video acquirer 10001) may secure point cloud data from information (e.g., depth information, color information, etc.) related to the acquisition process of the point cloud video.
The point cloud content providing system (for example, the transmission device 10000 or the point cloud video encoder 10002) according to the embodiments may encode the point cloud data (20001). The point cloud content providing system may encode the point cloud data based on point cloud compression coding. As described above, the point cloud data may include the geometry and attributes of a point. Accordingly, the point cloud content providing system may perform geometry encoding of encoding the geometry and output a geometry bitstream. The point cloud content providing system may perform attribute encoding of encoding attributes and output an attribute bitstream. According to embodiments, the point cloud content providing system may perform the attribute encoding based on the geometry encoding. The geometry bitstream and the attribute bitstream according to the embodiments may be multiplexed and output as one bitstream. The bitstream according to the embodiments may further contain signaling information related to the geometry encoding and attribute encoding.
The point cloud content providing system (for example, the transmission device 10000 or the transmitter 10003) according to the embodiments may transmit the encoded point cloud data (20002). As illustrated in
The point cloud content providing system (for example, the reception device 10004 or the receiver 10005) according to the embodiments may receive the bitstream containing the encoded point cloud data. In addition, the point cloud content providing system (for example, the reception device 10004 or the receiver 10005) may demultiplex the bitstream.
The point cloud content providing system (e.g., the reception device 10004 or the point cloud video decoder 10005) may decode the encoded point cloud data (e.g., the geometry bitstream, the attribute bitstream) transmitted in the bitstream. The point cloud content providing system (for example, the reception device 10004 or the point cloud video decoder 10005) may decode the point cloud video data based on the signaling information related to encoding of the point cloud video data contained in the bitstream. The point cloud content providing system (for example, the reception device 10004 or the point cloud video decoder 10005) may decode the geometry bitstream to reconstruct the positions (geometry) of points. The point cloud content providing system may reconstruct the attributes of the points by decoding the attribute bitstream based on the reconstructed geometry. The point cloud content providing system (for example, the reception device 10004 or the point cloud video decoder 10005) may reconstruct the point cloud video based on the positions according to the reconstructed geometry and the decoded attributes.
The point cloud content providing system according to the embodiments (for example, the reception device 10004 or the renderer 10007) may render the decoded point cloud data (20004). The point cloud content providing system (for example, the reception device 10004 or the renderer 10007) may render the geometry and attributes decoded through the decoding process, using various rendering methods. Points in the point cloud content may be rendered to a vertex having a certain thickness, a cube having a specific minimum size centered on the corresponding vertex position, or a circle centered on the corresponding vertex position. All or part of the rendered point cloud content is provided to the user through a display (e.g., a VR/AR display, a general display, etc.).
The point cloud content providing system (for example, the reception device 10004) according to the embodiments may secure feedback information (20005). The point cloud content providing system may encode and/or decode point cloud data based on the feedback information. The feedback information and the operation of the point cloud content providing system according to the embodiments are the same as the feedback information and the operation described with reference to
Point cloud content includes a point cloud video (images and/or videos) representing an object and/or environment located in various 3D spaces (e.g., a 3D space representing a real environment, a 3D space representing a virtual environment, etc.). Accordingly, the point cloud content providing system according to the embodiments may capture a point cloud video using one or more cameras (e.g., an infrared camera capable of securing depth information, an RGB camera capable of extracting color information corresponding to the depth information, etc.), a projector (e.g., an infrared pattern projector to secure depth information), a LiDAR, or the like. The point cloud content providing system according to the embodiments may extract the shape of geometry composed of points in a 3D space from the depth information and extract the attributes of each point from the color information to secure point cloud data. An image and/or video according to the embodiments may be captured based on at least one of the inward-facing technique and the outward-facing technique.
The left part of
The right part of
As shown in the figure, the point cloud content may be generated based on the capturing operation of one or more cameras. In this case, the coordinate system may differ among the cameras, and accordingly the point cloud content providing system may calibrate one or more cameras to set a global coordinate system before the capturing operation. In addition, the point cloud content providing system may generate point cloud content by synthesizing an arbitrary image and/or video with an image and/or video captured by the above-described capture technique. The point cloud content providing system may not perform the capturing operation described in
The point cloud content providing system may generate one piece of point cloud content by performing coordinate transformation on points of the point cloud video secured from each camera. The point cloud content providing system may perform coordinate transformation on the points based on the coordinates of the position of each camera. Accordingly, the point cloud content providing system may generate content representing one wide range, or may generate point cloud content having a high density of points.
As described with reference to
The point cloud encoder according to the embodiments includes a coordinate transformer (Transform coordinates) 40000, a quantizer (Quantize and remove points (voxelize)) 40001, an octree analyzer (Analyze octree) 40002, and a surface approximation analyzer (Analyze surface approximation) 40003, an arithmetic encoder (Arithmetic encode) 40004, a geometric reconstructor (Reconstruct geometry) 40005, a color transformer (Transform colors) 40006, an attribute transformer (Transform attributes) 40007, a RAHT transformer (RAHT) 40008, an LOD generator (Generate LOD) 40009, a lifting transformer (Lifting) 40010, a coefficient quantizer (Quantize coefficients) 40011, and/or an arithmetic encoder (Arithmetic encode) 40012.
The coordinate transformer 40000, the quantizer 40001, the octree analyzer 40002, the surface approximation analyzer 40003, the arithmetic encoder 40004, and the geometry reconstructor 40005 may perform geometry encoding. The geometry encoding according to the embodiments may include octree geometry coding, direct coding, trisoup geometry encoding, and entropy encoding. The direct coding and trisoup geometry encoding are applied selectively or in combination. The geometry encoding is not limited to the above-described example.
As shown in the figure, the coordinate transformer 40000 according to the embodiments receives positions and transforms the same into coordinates. For example, the positions may be transformed into position information in a three-dimensional space (for example, a three-dimensional space represented by an XYZ coordinate system). The position information in the three-dimensional space according to the embodiments may be referred to as geometry information.
The quantizer 40001 according to the embodiments quantizes the geometry. For example, the quantizer 40001 may quantize the points based on a minimum position value of all points (for example, a minimum value on each of the X, Y, and Z axes). The quantizer 40001 performs a quantization operation of multiplying the difference between the minimum position value and the position value of each point by a preset quantization scale value and then finding the nearest integer value by rounding the value obtained through the multiplication. Thus, one or more points may have the same quantized position (or position value). The quantizer 40001 according to the embodiments performs voxelization based on the quantized positions to reconstruct quantized points. As in the case of a pixel, which is the minimum unit containing 2D image/video information, points of point cloud content (or 3D point cloud video) according to the embodiments may be included in one or more voxels. The term voxel, which is a compound of volume and pixel, refers to a 3D cubic space generated when a 3D space is divided into units (unit=1.0) based on the axes representing the 3D space (e.g., X-axis, Y-axis, and Z-axis). The quantizer 40001 may match groups of points in the 3D space with voxels. According to embodiments, one voxel may include only one point. According to embodiments, one voxel may include one or more points. In order to express one voxel as one point, the position of the center of a voxel may be set based on the positions of one or more points included in the voxel. In this case, attributes of all positions included in one voxel may be combined and assigned to the voxel.
The octree analyzer 40002 according to the embodiments performs octree geometry coding (or octree coding) to present voxels in an octree structure. The octree structure represents points matched with voxels, based on the octal tree structure.
The surface approximation analyzer 40003 according to the embodiments may analyze and approximate the octree. The octree analysis and approximation according to the embodiments is a process of analyzing a region containing a plurality of points to efficiently provide octree and voxelization.
The arithmetic encoder 40004 according to the embodiments performs entropy encoding on the octree and/or the approximated octree. For example, the encoding scheme includes arithmetic encoding. As a result of the encoding, a geometry bitstream is generated.
The color transformer 40006, the attribute transformer 40007, the RAHT transformer 40008, the LOD generator 40009, the lifting transformer 40010, the coefficient quantizer 40011, and/or the arithmetic encoder 40012 perform attribute encoding. As described above, one point may have one or more attributes. The attribute encoding according to the embodiments is equally applied to the attributes that one point has. However, when an attribute (e.g., color) includes one or more elements, attribute encoding is independently applied to each element. The attribute encoding according to the embodiments includes color transform coding, attribute transform coding, region adaptive hierarchical transform (RAHT) coding, interpolation-based hierarchical nearest-neighbor prediction (prediction transform) coding, and interpolation-based hierarchical nearest-neighbor prediction with an update/lifting step (lifting transform) coding. Depending on the point cloud content, the RAHT coding, the prediction transform coding and the lifting transform coding described above may be selectively used, or a combination of one or more of the coding schemes may be used. The attribute encoding according to the embodiments is not limited to the above-described example.
The color transformer 40006 according to the embodiments performs color transform coding of transforming color values (or textures) included in the attributes. For example, the color transformer 40006 may transform the format of color information (for example, from RGB to YCbCr). The operation of the color transformer 40006 according to embodiments may be optionally applied according to the color values included in the attributes.
The geometry reconstructor 40005 according to the embodiments reconstructs (decompresses) the octree and/or the approximated octree. The geometry reconstructor 40005 reconstructs the octree/voxels based on the result of analyzing the distribution of points. The reconstructed octree/voxels may be referred to as reconstructed geometry (restored geometry).
The attribute transformer 40007 according to the embodiments performs attribute transformation to transform the attributes based on the reconstructed geometry and/or the positions on which geometry encoding is not performed. As described above, since the attributes are dependent on the geometry, the attribute transformer 40007 may transform the attributes based on the reconstructed geometry information. For example, based on the position value of a point included in a voxel, the attribute transformer 40007 may transform the attribute of the point at the position. As described above, when the position of the center of a voxel is set based on the positions of one or more points included in the voxel, the attribute transformer 40007 transforms the attributes of the one or more points. When the trisoup geometry encoding is performed, the attribute transformer 40007 may transform the attributes based on the trisoup geometry encoding.
The attribute transformer 40007 may perform the attribute transformation by calculating the average of attributes or attribute values of neighboring points (e.g., color or reflectance of each point) within a specific position/radius from the position (or position value) of the center of each voxel. The attribute transformer 40007 may apply a weight according to the distance from the center to each point in calculating the average. Accordingly, each voxel has a position and a calculated attribute (or attribute value).
The attribute transformer 40007 may search for neighboring points existing within a specific position/radius from the position of the center of each voxel based on the K-D tree or the Morton code. The K-D tree is a binary search tree and supports a data structure capable of managing points based on the positions such that nearest neighbor search (NNS) can be performed quickly. The Morton code is generated by presenting coordinates (e.g., (x, y, z)) representing 3D positions of all points as bit values and mixing the bits. For example, when the coordinates representing the position of a point are (5, 9, 1), the bit values for the coordinates are (0101, 1001, 0001). Mixing the bit values according to the bit index in order of z, y, and x yields 010001000111. This value is expressed as a decimal number of 1095. That is, the Morton code value of the point having coordinates (5, 9, 1) is 1095. The attribute transformer 40007 may order the points based on the Morton code values and perform NNS through a depth-first traversal process. After the attribute transformation operation, the K-D tree or the Morton code is used when the NNS is needed in another transformation process for attribute coding.
As shown in the figure, the transformed attributes are input to the RAHT transformer 40008 and/or the LOD generator 40009.
The RAHT transformer 40008 according to the embodiments performs RAHT coding for predicting attribute information based on the reconstructed geometry information. For example, the RAHT transformer 40008 may predict attribute information of a node at a higher level in the octree based on the attribute information associated with a node at a lower level in the octree.
The LOD generator 40009 according to the embodiments generates a level of detail (LOD) to perform prediction transform coding. The LOD according to the embodiments is a degree of detail of point cloud content. As the LOD value decrease, it indicates that the detail of the point cloud content is degraded. As the LOD value increases, it indicates that the detail of the point cloud content is enhanced. Points may be classified by the LOD.
The lifting transformer 40010 according to the embodiments performs lifting transform coding of transforming the attributes a point cloud based on weights. As described above, lifting transform coding may be optionally applied.
The coefficient quantizer 40011 according to the embodiments quantizes the attribute-coded attributes based on coefficients.
The arithmetic encoder 40012 according to the embodiments encodes the quantized attributes based on arithmetic coding.
Although not shown in the figure, the elements of the point cloud encoder of
As described with reference to
The upper part of
d=Ceil(Log2(Max(xnint, ynint, znint, n=1, . . . , N)+1))
As shown in the middle of the upper part of
The lower part of
The point cloud encoder (for example, the point cloud encoder of
Accordingly, for the above-described specific region (or a node other than the leaf node of the octree), the point cloud encoder according to the embodiments may skip voxelization and perform direct coding to directly code the positions of points included in the specific region. The coordinates of a direct coding point according to the embodiments are referred to as direct coding mode (DCM). The point cloud encoder according to the embodiments may also perform trisoup geometry encoding, which is to reconstruct the positions of the points in the specific region (or node) based on voxels, based on a surface model. The trisoup geometry encoding is geometry encoding that represents an object as a series of triangular meshes. Accordingly, the point cloud decoder may generate a point cloud from the mesh surface. The direct coding and trisoup geometry encoding according to the embodiments may be selectively performed. In addition, the direct coding and trisoup geometry encoding according to the embodiments may be performed in combination with octree geometry coding (or octree coding).
To perform direct coding, the option to use the direct mode for applying direct coding should be activated. A node to which direct coding is to be applied is not a leaf node, and points less than a threshold should be present within a specific node. In addition, the total number of points to which direct coding is to be applied should not exceed a preset threshold. When the conditions above are satisfied, the point cloud encoder (or the arithmetic encoder 40004) according to the embodiments may perform entropy coding on the positions (or position values) of the points.
The point cloud encoder (for example, the surface approximation analyzer 40003) according to the embodiments may determine a specific level of the octree (a level less than the depth d of the octree), and the surface model may be used staring with that level to perform trisoup geometry encoding to reconstruct the positions of points in the region of the node based on voxels (Trisoup mode). The point cloud encoder according to the embodiments may specify a level at which trisoup geometry encoding is to be applied. For example, when the specific level is equal to the depth of the octree, the point cloud encoder does not operate in the trisoup mode. In other words, the point cloud encoder according to the embodiments may operate in the trisoup mode only when the specified level is less than the value of depth of the octree. The 3D cube region of the nodes at the specified level according to the embodiments is called a block. One block may include one or more voxels. The block or voxel may correspond to a brick. Geometry is represented as a surface within each block. The surface according to embodiments may intersect with each edge of a block at most once.
One block has 12 edges, and accordingly there are at least 12 intersections in one block. Each intersection is called a vertex (or apex). A vertex present along an edge is detected when there is at least one occupied voxel adjacent to the edge among all blocks sharing the edge. The occupied voxel according to the embodiments refers to a voxel containing a point. The position of the vertex detected along the edge is the average position along the edge of all voxels adjacent to the edge among all blocks sharing the edge.
Once the vertex is detected, the point cloud encoder according to the embodiments may perform entropy encoding on the starting point (x, y, z) of the edge, the direction vector (Δx, Δy, Δz) of the edge, and the vertex position value (relative position value within the edge). When the trisoup geometry encoding is applied, the point cloud encoder according to the embodiments (for example, the geometry reconstructor 40005) may generate restored geometry (reconstructed geometry) by performing the triangle reconstruction, up-sampling, and voxelization processes.
The vertices positioned at the edge of the block determine a surface that passes through the block. The surface according to the embodiments is a non-planar polygon. In the triangle reconstruction process, a surface represented by a triangle is reconstructed based on the starting point of the edge, the direction vector of the edge, and the position values of the vertices. The triangle reconstruction process is performed by: i) calculating the centroid value of each vertex, ii) subtracting the center value from each vertex value, and iii) estimating the sum of the squares of the values obtained by the subtraction.
The minimum value of the sum is estimated, and the projection process is performed according to the axis with the minimum value. For example, when the element x is the minimum, each vertex is projected on the x-axis with respect to the center of the block, and projected on the (y, z) plane. When the values obtained through projection on the (y, z) plane are (ai, bi), the value of θ is estimated through atan2(bi, ai), and the vertices are ordered based on the value of θ. The table below shows a combination of vertices for creating a triangle according to the number of the vertices. The vertices are ordered from 1 to n. The table below shows that for four vertices, two triangles may be constructed according to combinations of vertices. The first triangle may consist of vertices 1, 2, and 3 among the ordered vertices, and the second triangle may consist of vertices 3, 4, and 1 among the ordered vertices.
The upsampling process is performed to add points in the middle along the edge of the triangle and perform voxelization. The added points are generated based on the upsampling factor and the width of the block. The added points are called refined vertices. The point cloud encoder according to the embodiments may voxelize the refined vertices. In addition, the point cloud encoder may perform attribute encoding based on the voxelized positions (or position values).
In order to increase the compression efficiency of the point cloud video, the point cloud encoder according to the embodiments may perform entropy coding based on context adaptive arithmetic coding.
As described with reference to
The right part of
As described with reference to
The point cloud encoder (for example, the LOD generator 40009) may classify (reorganize) points by LOD. The figure shows the point cloud content corresponding to LODs. The leftmost picture in the figure represents original point cloud content. The second picture from the left of the figure represents distribution of the points in the lowest LOD, and the rightmost picture in the figure represents distribution of the points in the highest LOD. That is, the points in the lowest LOD are sparsely distributed, and the points in the highest LOD are densely distributed. That is, as the LOD rises in the direction pointed by the arrow indicated at the bottom of the figure, the space (or distance) between points is narrowed.
As described with reference to
The upper part of
As described with reference to
The point cloud encoder according to the embodiments may generate a predictor for points to perform prediction transform coding for setting a predicted attribute (or predicted attribute value) of each point. That is, N predictors may be generated for N points. The predictor according to the embodiments may calculate a weight (=1/distance) based on the LOD value of each point, indexing information about neighboring points present within a set distance for each LOD, and a distance to the neighboring points.
The predicted attribute (or attribute value) according to the embodiments is set to the average of values obtained by multiplying the attributes (or attribute values) (e.g., color, reflectance, etc.) of neighbor points set in the predictor of each point by a weight (or weight value) calculated based on the distance to each neighbor point. The point cloud encoder according to the embodiments (for example, the coefficient quantizer 40011) may quantize and inversely quantize the residuals (which may be called residual attributes, residual attribute values, or attribute prediction residuals) obtained by subtracting a predicted attribute (attribute value) from the attribute (attribute value) of each point. The quantization process is configured as shown in the following table.
When the predictor of each point has neighbor points, the point cloud encoder (e.g., the arithmetic encoder 40012) according to the embodiments may perform entropy coding on the quantized and inversely quantized residual values as described above. When the predictor of each point has no neighbor point, the point cloud encoder according to the embodiments (for example, the arithmetic encoder 40012) may perform entropy coding on the attributes of the corresponding point without performing the above-described operation.
The point cloud encoder according to the embodiments (for example, the lifting transformer 40010) may generate a predictor of each point, set the calculated LOD and register neighbor points in the predictor, and set weights according to the distances to neighbor points to perform lifting transform coding. The lifting transform coding according to the embodiments is similar to the above-described prediction transform coding, but differs therefrom in that weights are cumulatively applied to attribute values. The process of cumulatively applying weights to the attribute values according to embodiments is configured as follows.
1) Create an array Quantization Weight (QW) for storing the weight value of each point. The initial value of all elements of QW is 1.0. Multiply the QW values of the predictor indexes of the neighbor nodes registered in the predictor by the weight of the predictor of the current point, and add the values obtained by the multiplication.
2) Lift prediction process: Subtract the value obtained by multiplying the attribute value of the point by the weight from the existing attribute value to calculate a predicted attribute value.
3) Create temporary arrays called updateweight and update and initialize the temporary arrays to zero.
4) Cumulatively add the weights calculated by multiplying the weights calculated for all predictors by a weight stored in the QW corresponding to a predictor index to the updateweight array as indexes of neighbor nodes. Cumulatively add, to the update array, a value obtained by multiplying the attribute value of the index of a neighbor node by the calculated weight.
5) Lift update process: Divide the attribute values of the update array for all predictors by the weight value of the updateweight array of the predictor index, and add the existing attribute value to the values obtained by the division.
6) Calculate predicted attributes by multiplying the attribute values updated through the lift update process by the weight updated through the lift prediction process (stored in the QW) for all predictors. The point cloud encoder (e.g., coefficient quantizer 40011) according to the embodiments quantizes the predicted attribute values. In addition, the point cloud encoder (e.g., the arithmetic encoder 40012) performs entropy coding on the quantized attribute values.
The point cloud encoder (for example, the RAHT transformer 40008) according to the embodiments may perform RAHT transform coding in which attributes of nodes of a higher level are predicted using the attributes associated with nodes of a lower level in the octree. RAHT transform coding is an example of attribute intra coding through an octree backward scan. The point cloud encoder according to the embodiments scans the entire region from the voxel and repeats the merging process of merging the voxels into a larger block at each step until the root node is reached. The merging process according to the embodiments is performed only on the occupied nodes. The merging process is not performed on the empty node. The merging process is performed on an upper node immediately above the empty node.
The equation below represents a RAHT transformation matrix. In the equation, gι
Here, gι−1
The point cloud decoder illustrated in
The point cloud decoder illustrated in
As described with reference to
The point cloud decoder according to the embodiments includes an arithmetic decoder (Arithmetic decode) 11000, an octree synthesizer (Synthesize octree) 11001, a surface approximation synthesizer (Synthesize surface approximation) 11002, and a geometry reconstructor (Reconstruct geometry) 11003, a coordinate inverse transformer (Inverse transform coordinates) 11004, an arithmetic decoder (Arithmetic decode) 11005, an inverse quantizer (Inverse quantize) 11006, a RAHT transformer 11007, an LOD generator (Generate LOD) 11008, an inverse lifter (inverse lifting) 11009, and/or a color inverse transformer (Inverse transform colors) 11010.
The arithmetic decoder 11000, the octree synthesizer 11001, the surface approximation synthesizer 11002, and the geometry reconstructor 11003, and the coordinate inverse transformer 11004 may perform geometry decoding. The geometry decoding according to the embodiments may include direct coding and trisoup geometry decoding. The direct coding and trisoup geometry decoding are selectively applied. The geometry decoding is not limited to the above-described example, and is performed as an inverse process of the geometry encoding described with reference to
The arithmetic decoder 11000 according to the embodiments decodes the received geometry bitstream based on the arithmetic coding. The operation of the arithmetic decoder 11000 corresponds to the inverse process of the arithmetic encoder 40004.
The octree synthesizer 11001 according to the embodiments may generate an octree by acquiring an occupancy code from the decoded geometry bitstream (or information on the geometry secured as a result of decoding). The occupancy code is configured as described in detail with reference to
When the trisoup geometry encoding is applied, the surface approximation synthesizer 11002 according to the embodiments may synthesize a surface based on the decoded geometry and/or the generated octree.
The geometry reconstructor 11003 according to the embodiments may regenerate geometry based on the surface and/or the decoded geometry. As described with reference to
The coordinate inverse transformer 11004 according to the embodiments may acquire positions of the points by transforming the coordinates based on the reconstructed geometry.
The arithmetic decoder 11005, the inverse quantizer 11006, the RAHT transformer 11007, the LOD generator 11008, the inverse lifter 11009, and/or the color inverse transformer 11010 may perform the attribute decoding described with reference to
The arithmetic decoder 11005 according to the embodiments decodes the attribute bitstream by arithmetic coding.
The inverse quantizer 11006 according to the embodiments inversely quantizes the information about the decoded attribute bitstream or attributes secured as a result of the decoding, and outputs the inversely quantized attributes (or attribute values). The inverse quantization may be selectively applied based on the attribute encoding of the point cloud encoder.
According to embodiments, the RAHT transformer 11007, the LOD generator 11008, and/or the inverse lifter 11009 may process the reconstructed geometry and the inversely quantized attributes. As described above, the RAHT transformer 11007, the LOD generator 11008, and/or the inverse lifter 11009 may selectively perform a decoding operation corresponding to the encoding of the point cloud encoder.
The color inverse transformer 11010 according to the embodiments performs inverse transform coding to inversely transform a color value (or texture) included in the decoded attributes. The operation of the color inverse transformer 11010 may be selectively performed based on the operation of the color transformer 40006 of the point cloud encoder.
Although not shown in the figure, the elements of the point cloud decoder of
The transmission device shown in
The data input unit 12000 according to the embodiments receives or acquires point cloud data. The data input unit 12000 may perform an operation and/or acquisition method the same as or similar to the operation and/or acquisition method of the point cloud video acquirer 10001 (or the acquisition process 20000 described with reference to
The data input unit 12000, the quantization processor 12001, the voxelization processor 12002, the octree occupancy code generator 12003, the surface model processor 12004, the intra/inter-coding processor 12005, and the arithmetic coder 12006 perform geometry encoding. The geometry encoding according to the embodiments is the same as or similar to the geometry encoding described with reference to
The quantization processor 12001 according to the embodiments quantizes geometry (e.g., position values of points). The operation and/or quantization of the quantization processor 12001 is the same as or similar to the operation and/or quantization of the quantizer 40001 described with reference to
The voxelization processor 12002 according to the embodiments voxelizes the quantized position values of the points. The voxelization processor 120002 may perform an operation and/or process the same or similar to the operation and/or the voxelization process of the quantizer 40001 described with reference to
The octree occupancy code generator 12003 according to the embodiments performs octree coding on the voxelized positions of the points based on an octree structure. The octree occupancy code generator 12003 may generate an occupancy code. The octree occupancy code generator 12003 may perform an operation and/or method the same as or similar to the operation and/or method of the point cloud encoder (or the octree analyzer 40002) described with reference to
The surface model processor 12004 according to the embodiments may perform trigsoup geometry encoding based on a surface model to reconstruct the positions of points in a specific region (or node) on a voxel basis. The surface model processor 12004 may perform an operation and/or method the same as or similar to the operation and/or method of the point cloud encoder (for example, the surface approximation analyzer 40003) described with reference to
The intra/inter-coding processor 12005 according to the embodiments may perform intra/inter-coding on point cloud data. The intra/inter-coding processor 12005 may perform coding the same as or similar to the intra/inter-coding described with reference to
The arithmetic coder 12006 according to the embodiments performs entropy encoding on an octree of the point cloud data and/or an approximated octree. For example, the encoding scheme includes arithmetic encoding. The arithmetic coder 12006 performs an operation and/or method the same as or similar to the operation and/or method of the arithmetic encoder 40004.
The metadata processor 12007 according to the embodiments processes metadata about the point cloud data, for example, a set value, and provides the same to a necessary processing process such as geometry encoding and/or attribute encoding. Also, the metadata processor 12007 according to the embodiments may generate and/or process signaling information related to the geometry encoding and/or the attribute encoding. The signaling information according to the embodiments may be encoded separately from the geometry encoding and/or the attribute encoding. The signaling information according to the embodiments may be interleaved.
The color transform processor 12008, the attribute transform processor 12009, the prediction/lifting/RAHT transform processor 12010, and the arithmetic coder 12011 perform the attribute encoding. The attribute encoding according to the embodiments is the same as or similar to the attribute encoding described with reference to
The color transform processor 12008 according to the embodiments performs color transform coding to transform color values included in attributes. The color transform processor 12008 may perform color transform coding based on the reconstructed geometry. The reconstructed geometry is the same as described with reference to
The attribute transform processor 12009 according to the embodiments performs attribute transformation to transform the attributes based on the reconstructed geometry and/or the positions on which geometry encoding is not performed. The attribute transform processor 12009 performs an operation and/or method the same as or similar to the operation and/or method of the attribute transformer 40007 described with reference to
The arithmetic coder 12011 according to the embodiments may encode the coded attributes based on the arithmetic coding. The arithmetic coder 12011 performs an operation and/or method the same as or similar to the operation and/or method of the arithmetic encoder 400012.
The transmission processor 12012 according to the embodiments may transmit each bitstream containing encoded geometry and/or encoded attributes and metadata information, or transmit one bitstream configured with the encoded geometry and/or the encoded attributes and the metadata information. When the encoded geometry and/or the encoded attributes and the metadata information according to the embodiments are configured into one bitstream, the bitstream may include one or more sub-bitstreams. The bitstream according to the embodiments may contain signaling information including a sequence parameter set (SPS) for signaling of a sequence level, a geometry parameter set (GPS) for signaling of geometry information coding, an attribute parameter set (APS) for signaling of attribute information coding, and a tile parameter set (TPS) for signaling of a tile level, and slice data. The slice data may include information about one or more slices. One slice according to embodiments may include one geometry bitstream Geom00 and one or more attribute bitstreams Attr00 and Attr10.
A slice refers to a series of syntax elements representing the entirety or part of a coded point cloud frame.
The TPS according to the embodiments may include information about each tile (for example, coordinate information and height/size information about a bounding box) for one or more tiles. The geometry bitstream may contain a header and a payload. The header of the geometry bitstream according to the embodiments may contain a parameter set identifier (geom_parameter_set_id), a tile identifier (geom_tile_id) and a slice identifier (geom_slice_id) included in the GPS, and information about the data contained in the payload. As described above, the metadata processor 12007 according to the embodiments may generate and/or process the signaling information and transmit the same to the transmission processor 12012. According to embodiments, the elements to perform geometry encoding and the elements to perform attribute encoding may share data/information with each other as indicated by dotted lines. The transmission processor 12012 according to the embodiments may perform an operation and/or transmission method the same as or similar to the operation and/or transmission method of the transmitter 10003. Details are the same as those described with reference to
The reception device illustrated in
The reception device according to the embodiment includes a receiver 13000, a reception processor 13001, an arithmetic decoder 13002, an occupancy code-based octree reconstruction processor 13003, a surface model processor (triangle reconstruction, up-sampling, voxelization) 13004, an inverse quantization processor 13005, a metadata parser 13006, an arithmetic decoder 13007, an inverse quantization processor 13008, a prediction/lifting/RAHT inverse transform processor 13009, a color inverse transform processor 13010, and/or a renderer 13011. Each element for decoding according to the embodiments may perform an inverse process of the operation of a corresponding element for encoding according to the embodiments.
The receiver 13000 according to the embodiments receives point cloud data. The receiver 13000 may perform an operation and/or reception method the same as or similar to the operation and/or reception method of the receiver 10005 of
The reception processor 13001 according to the embodiments may acquire a geometry bitstream and/or an attribute bitstream from the received data. The reception processor 13001 may be included in the receiver 13000.
The arithmetic decoder 13002, the occupancy code-based octree reconstruction processor 13003, the surface model processor 13004, and the inverse quantization processor 13005 may perform geometry decoding. The geometry decoding according to embodiments is the same as or similar to the geometry decoding described with reference to
The arithmetic decoder 13002 according to the embodiments may decode the geometry bitstream based on arithmetic coding. The arithmetic decoder 13002 performs an operation and/or coding the same as or similar to the operation and/or coding of the arithmetic decoder 11000.
The occupancy code-based octree reconstruction processor 13003 according to the embodiments may reconstruct an octree by acquiring an occupancy code from the decoded geometry bitstream (or information about the geometry secured as a result of decoding). The occupancy code-based octree reconstruction processor 13003 performs an operation and/or method the same as or similar to the operation and/or octree generation method of the octree synthesizer 11001. When the trisoup geometry encoding is applied, the surface model processor 13004 according to the embodiments may perform trisoup geometry decoding and related geometry reconstruction (for example, triangle reconstruction, up-sampling, voxelization) based on the surface model method. The surface model processor 13004 performs an operation the same as or similar to that of the surface approximation synthesizer 11002 and/or the geometry reconstructor 11003.
The inverse quantization processor 13005 according to the embodiments may inversely quantize the decoded geometry.
The metadata parser 13006 according to the embodiments may parse metadata contained in the received point cloud data, for example, a set value. The metadata parser 13006 may pass the metadata to geometry decoding and/or attribute decoding. The metadata is the same as that described with reference to
The arithmetic decoder 13007, the inverse quantization processor 13008, the prediction/lifting/RAHT inverse transform processor 13009 and the color inverse transform processor 13010 perform attribute decoding. The attribute decoding is the same as or similar to the attribute decoding described with reference to
The arithmetic decoder 13007 according to the embodiments may decode the attribute bitstream by arithmetic coding. The arithmetic decoder 13007 may decode the attribute bitstream based on the reconstructed geometry. The arithmetic decoder 13007 performs an operation and/or coding the same as or similar to the operation and/or coding of the arithmetic decoder 11005.
The inverse quantization processor 13008 according to the embodiments may inversely quantize the decoded attribute bitstream. The inverse quantization processor 13008 performs an operation and/or method the same as or similar to the operation and/or inverse quantization method of the inverse quantizer 11006.
The prediction/lifting/RAHT inverse transformer 13009 according to the embodiments may process the reconstructed geometry and the inversely quantized attributes. The prediction/lifting/RAHT inverse transform processor 13009 performs one or more of operations and/or decoding the same as or similar to the operations and/or decoding of the RAHT transformer 11007, the LOD generator 11008, and/or the inverse lifter 11009. The color inverse transform processor 13010 according to the embodiments performs inverse transform coding to inversely transform color values (or textures) included in the decoded attributes. The color inverse transform processor 13010 performs an operation and/or inverse transform coding the same as or similar to the operation and/or inverse transform coding of the color inverse transformer 11010. The renderer 13011 according to the embodiments may render the point cloud data.
The structure of
The cloud network 1400 may represent a network that constitutes part of the cloud computing infrastructure or is present in the cloud computing infrastructure. Here, the cloud network 1400 may be configured using a 3G network, 4G or Long Term Evolution (LTE) network, or a 5G network.
The server 1460 may be connected to at least one of the robot 1410, the self-driving vehicle 1420, the XR device 1430, the smartphone 1440, the home appliance 1450, and/or the HMD 1470 over the cloud network 1400 and may assist in at least a part of the processing of the connected devices 1410 to 1470.
The HMD 1470 represents one of the implementation types of the XR device and/or the PCC device according to the embodiments. The HMD type device according to the embodiments includes a communication unit, a control unit, a memory, an I/O unit, a sensor unit, and a power supply unit.
Hereinafter, various embodiments of the devices 1410 to 1450 to which the above-described technology is applied will be described. The devices 1410 to 1450 illustrated in
<PCC+XR>
The XR/PCC device 1430 may employ PCC technology and/or XR (AR+VR) technology, and may be implemented as an HMD, a head-up display (HUD) provided in a vehicle, a television, a mobile phone, a smartphone, a computer, a wearable device, a home appliance, a digital signage, a vehicle, a stationary robot, or a mobile robot.
The XR/PCC device 1430 may analyze 3D point cloud data or image data acquired through various sensors or from an external device and generate position data and attribute data about 3D points. Thereby, the XR/PCC device 1430 may acquire information about the surrounding space or a real object, and render and output an XR object. For example, the XR/PCC device 1430 may match an XR object including auxiliary information about a recognized object with the recognized object and output the matched XR object.
<PCC+XR+Mobile Phone>
The XR/PCC device 1430 may be implemented as a mobile phone 1440 by applying PCC technology.
The mobile phone 1440 may decode and display point cloud content based on the PCC technology.
<PCC+Self-Driving+XR>
The self-driving vehicle 1420 may be implemented as a mobile robot, a vehicle, an unmanned aerial vehicle, or the like by applying the PCC technology and the XR technology.
The self-driving vehicle 1420 to which the XR/PCC technology is applied may represent a self-driving vehicle provided with means for providing an XR image, or a self-driving vehicle that is a target of control/interaction in the XR image. In particular, the self-driving vehicle 1420 which is a target of control/interaction in the XR image may be distinguished from the XR device 1430 and may be operatively connected thereto.
The self-driving vehicle 1420 having means for providing an XR/PCC image may acquire sensor information from sensors including a camera, and output the generated XR/PCC image based on the acquired sensor information. For example, the self-driving vehicle 1420 may have an HUD and output an XR/PCC image thereto, thereby providing an occupant with an XR/PCC object corresponding to a real object or an object present on the screen.
When the XR/PCC object is output to the HUD, at least a part of the XR/PCC object may be output to overlap the real object to which the occupant's eyes are directed. On the other hand, when the XR/PCC object is output on a display provided inside the self-driving vehicle, at least a part of the XR/PCC object may be output to overlap an object on the screen. For example, the self-driving vehicle 1220 may output XR/PCC objects corresponding to objects such as a road, another vehicle, a traffic light, a traffic sign, a two-wheeled vehicle, a pedestrian, and a building.
The virtual reality (VR) technology, the augmented reality (AR) technology, the mixed reality (MR) technology and/or the point cloud compression (PCC) technology according to the embodiments are applicable to various devices.
In other words, the VR technology is a display technology that provides only CG images of real-world objects, backgrounds, and the like. On the other hand, the AR technology refers to a technology that shows a virtually created CG image on the image of a real object. The MR technology is similar to the AR technology described above in that virtual objects to be shown are mixed and combined with the real world. However, the MR technology differs from the AR technology in that the AR technology makes a clear distinction between a real object and a virtual object created as a CG image and uses virtual objects as complementary objects for real objects, whereas the MR technology treats virtual objects as objects having equivalent characteristics as real objects. More specifically, an example of MR technology applications is a hologram service.
Recently, the VR, AR, and MR technologies are sometimes referred to as extended reality (XR) technology rather than being clearly distinguished from each other. Accordingly, embodiments of the present disclosure are applicable to any of the VR, AR, MR, and XR technologies. The encoding/decoding based on PCC, V-PCC, and G-PCC techniques is applicable to such technologies.
The PCC method/device according to the embodiments may be applied to a vehicle that provides a self-driving service.
A vehicle that provides the self-driving service is connected to a PCC device for wired/wireless communication.
When the point cloud data (PCC) transmission/reception device according to the embodiments is connected to a vehicle for wired/wireless communication, the device may receive/process content data related to an AR/VR/PCC service, which may be provided together with the self-driving service, and transmit the same to the vehicle. In the case where the PCC transmission/reception device is mounted on a vehicle, the PCC transmission/reception device may receive/process content data related to the AR/VR/PCC service according to a user input signal input through a user interface device and provide the same to the user. The vehicle or the user interface device according to the embodiments may receive a user input signal. The user input signal according to the embodiments may include a signal indicating the self-driving service.
The point cloud data transmission method/device according to embodiments may be construed as a term referring to the transmission device 10000 of
The point cloud data reception method/device according to embodiments may be construed as a term referring to the reception device 10004 of
The method/device for transmitting or receiving point cloud data according to the embodiments may be referred to simply as a method/device.
According to embodiments, geometry data, geometry information, and position information constituting point cloud data are to be construed as having the same meaning. Attribute data, attribute information, and attribute information constituting the point cloud data are to be construed as having the same meaning.
The method/device according to the embodiments may encode (compress) and decode (reconstruct) attributes of point cloud data based on a predictive tree. This scheme means prediction-based coding. The attribute coding may be performed based on the predictive tree structure used for predictive coding, and a compression and reconstruction method suitable for low latency applications may be provided.
Referring to
For example, in compressing position information, an octree-based compression method is used, and then attribute information is compressed based on the compressed data. In this case, predictive coding may be applied as a new compression technique for compressing the position information.
For point cloud data, predictive coding for predicting a point based on neighbor point information may be used. In the prediction-based method, the coding speed may be fast. However, when the attribute coding is used, all information about the point positions is required, and accordingly a delay may occur due to the attribute coding. For this reason, there is a limitation in fully utilizing the advantages of the prediction-based position coding.
In this regard, the method/device according to the embodiments may carry out an efficient predictive attribute coding scheme.
For example, the methods for prediction tree-based attribute coding according to the embodiments may include 1) using the same prediction mode as the position coding, 2) using a separate attribute prediction mode based on a parent-child relationship, and/or 3) predicting an attribute using a separate parent. Each operation will be described below.
The prediction tree-based point cloud attribute compression by the method/device according to the embodiments is a prediction-based compression technique suitable for use in a low-latency environment. The delay element present between the positional compression and the attribute compression may be removed, and attribute coding may be performed based on the predictive tree structure. The method/device according to the embodiments may maximize the low-latency effect of the prediction-based compression and may perform general point cloud attribute compression.
Also, a prediction tree generation method according to embodiments may be used for the prediction-based compression.
Also, the prediction-based attribute compression method according to the embodiments may be used independently of the position compression method (geometry coding).
The method/device for transmitting point cloud data according to the embodiments of
Specifically, for the input point cloud data, a) the compression target point cloud data may be sorted, b) a predictive tree structure may be generated, c) an attribute may be predicted based on the generated predictive tree structure, and d) a bitstream may be generated and transmitted by encoding a residual based on the predicted attribute.
The method/device for receiving point cloud data according to the embodiments of
Hereinafter, each operation according to the embodiments will be described.
A) Data Sorting and Compression Unit Configuration
The method/device according to the embodiments performs prediction by generating a predictive tree structure for point cloud data divided into predetermined units for prediction-based point cloud compression.
In this case, when the parent-child relationship in the predictive tree is close, the size of the residual generated through prediction is small because the parent point and the child point are similar to each other. That is, the prediction performance becomes excellent.
Therefore, before dividing the input point cloud data into compression units, similar points should be placed together in a compression unit.
For example, the points may be sorted in a Morton code order, points having similar radii in the cylindrical coordinate system may be grouped, or points having similar azimuth/elevation angles in the cylindrical coordinate system may be grouped. As similar points are grouped as described above, the compression performance may increase.
The method/device according to the embodiments performs compression by dividing the sorted point cloud data into predetermined compression units.
For the compression unit, the entire point cloud may be configured as a single compression unit, or locally adjacent points on XYZ coordinates may be grouped to configure a compression unit. Alternatively, the points may be sorted in a specific order and then divided in units of a certain number, may be configured in LoD units. Alternatively, the compression unit may be configured by grouping similar points in units of radius/azimuth/elevation in the cylindrical coordinate system.
For example, a compression unit may be configured in a unit of a certain number of points (e.g., 512, 1024, etc.). In a case where the compression unit is configured according to the input point cloud order and a case where Morton code-ordered points are used, a big difference may be made in compression efficiency even when the same predictive compression is used.
B) Generating a Predictive Tree Structure
A method/device according to embodiments may construct/generate a predictive tree in a compression unit.
A predictive tree construction method according to embodiments may include a predictive tree structure generation method for the prediction-based position information compression method. Accordingly, the same structure may be used for position compression (geometry encoding) and attribute compression (attribute encoding) in point cloud compression, and complexity may be reduced in a low-latency environment.
The predictive tree construction may establish a parent-child relationship between points nearest a specific point in position.
For example, when there are four points P0 P2 P4 P5 as shown in
In order to find neighbor points among points, a nearest neighbor search method based on the Euclidean distance may be used.
In this way, a predictive tree in which a neighbor relationship is indicated based on distance may be generated.
C) Predictive Tree Structure-Based Attribute Prediction
The method/device according to the embodiments may predict an attribute based on a parent-child relationship of the constructed (generated) predictive tree as follows: (1) performing attribute prediction based on position prediction; (2) using a separate attribute prediction mode different from the position prediction mode; (3) performing attribute prediction using a predictive tree structure.
(1) Performing attribute prediction based on position prediction
In performing the attribute prediction, the attribute prediction may be performed in the same way as the position prediction. For example, when V(p) is defined as the position of the p-th point, V(p−1) is defined as the position of the parent of V(p), and V(p−2) is defined as the position of the grandparent of V(p), the prediction error E may be defined as follows, and a method of minimizing E may be delivered as a position prediction mode.
1) E=[V(p)−a*V(p−1)−b];
2) E={[V(p)−V(p−1)]−a*[V(p−1)−V(p−2)]−b};
3) E={[V(p)+V(p−1)]/2−a*[V(p−1)+V(p−2)]/2−b};
4) E={[V(p)−V(p−1)]−a*[V(p−2)−V(p−3)]−b};
5) E={[V(p)+V(p−1)+V(p−2)]/3−a*[V(p−1)+V(p−2)+V(p−3)]/3−b};
6) E={[V(p)+V(p−1)+V(p−2)]/3−a′*[V(p−2)+V(p−3)+V(p−4)]/3−b′};
7) E={[V(p)−2V(p−1)+V(p−2)]−a*[V(p−1)−2V(p−2)+V(p−3)]−b}.
V′(p), which minimizes the error in the above equations, may be predicted as follows:
1) mode 1: V′(p)=a*V(p−1)+b;
2) mode 2: V′(p)=(a+1)*V(p−1)−a*V(p−2)+b;
3) mode 3: V′(p)=(a−1)*V(p−1)+a*V(p−2)+2b;
4) mode 4: V′(p)=V(p−1)+a*V(p−2)−a*V(p−3)+b
5) mode 5: V′(p)=(a−1)*V(p−1)+(a−1)*V(p−2)+a*V(p−3)+3b;
6) mode 6: V′(p)=V(p−1)+(a−1)*V(p−2)+a*V(p−3)+a*V(p−4)+3b;
7) mode 7: V′(p)=(a+2)*V(p−1)−(2a +1)*V(p−2)+a*V(p−3)+b.
For the above-disclosed various prediction methods, 1) a predetermined method may be used on the basis of a specific unit (e.g., slice, coding block, frame, N units, etc.), or information about the specific unit of the method used may be transmitted/received through signaling information, or 2) information on various prediction methods for a method of minimizing the error may be transmitted/received through signaling information for every point.
Also, for prediction coefficients a and b, 1) predetermined values may be used and signaled, or 2) coefficients for a method of minimizing the error may be used and signaled for every point, or 3) the coefficients may be defined as functions inversely proportional to the distance between points.
As such, the same mode is used for position compression, an operation defined in the mode (e.g., (a+1)A(p−1)−a(v(p−2)+b in mode 2) is predicted for each point.
That is, in the same way as for the position prediction, using the method of minimizing the error according to the equations for the current point, the parent point, the grandparent point, and the like, the attributes may be predicted according to the prediction mode based on the attribute values of the corresponding points: mode 1: A′(p)=a*A(p−1)+b, mode 2: A′(p)=a*A(p−1)+b, mode 2: A′(p)=(a+1)*A(p−1)−a*A(p−2)+b, mode 3: A′(p)=(a−1)*A(p−1)+a*A(p−2)+2b, mode 4: A′(p)=A(p−1)+a*A(p−2)−a*A(p−3)+b, mode 5: A′(p)=(a−1)*A(p−1)+(a−1)*A(p−2)+a*A(p−3)+3b.
In predicting attributes based on the position prediction according to the embodiments, an optimal mode for attribute compression is searched for based on the position information V(p). In the process of coding the attribute for the mode, an equation related to the attribute A(p) matching each position according to the embodiments may be used.
When a separate attribute prediction mode different from the position prediction mode according to the embodiments is used, prediction is performed based on the attribute A(p).
The two methods may differ from each other in terms of how to determine the mode and whether signaling is performed. For example, when the two methods designate mode 2, an attribute may be predicted using the equation corresponding to attribute coding mode 2 according to the embodiments, (mode 2: A′(p)=(a+1)*A(p−1)−a*A(p−2)+b).
When the same prediction method as the position prediction is used for attribute prediction as in the proposed method, the bitrate efficiency may be increased as much as the bits that should be used for the attribute prediction mode transmission by using the position prediction mode instead of delivering the attribute prediction mode.
Since the position prediction (geometry encoding) and the attribute prediction (attribute encoding) are performed in the same/similar manner, the method/device for receiving point cloud data according to the embodiments may efficiently perform position reconstruction (geometry decoding) and attribute reconstruction (attribute decoding).
(2) Using a separate attribute prediction mode different from the position prediction mode
In predicting an attribute, a prediction separate from the position prediction may be performed. For example, when A(p) is defined as the attribute of the p-th point, A(p−1) is defined as the attribute of the parent of the p-th point, and A(p−2) is defined as the attribute of the grandparent of the p-th point, the attribute prediction error E_a may be defined as follows, and the method of minimizing E_a may be separately delivered as an attribute prediction mode:
1) E_a=[A(p)−a*A (p−1)−b];
2) E_a={[A(p)−A(p−1)]−a*[A(p−1)−A(p−2)]−b};
3) E_a={[A(p)+A(p−1)]/2−a*[A(p−1)+A(p−2)]/2−b};
4) E_a={[A(p)−A(p−1)]−a*[A(p−2)−A(p−3)]−b};
5) E_a={[A(p)+A(p−1)+A(p−2)]/3−a*[A(p−1)+A(p−2)+A(p−3)]/3−b};
6) E_a={[A(p)+A(p−1)+A(p−2)]/3−a′*[A(p−2)+A(p−3)+A(p−4)]/3−b′};
7) E_a={[A(p)−2A(p−1)+A(p−2)]−a*[A(p−1)−2A(p−2)+A(p−3)]−b}.
A′(p), which minimizes the error in the above equations, may be predicted as follows:
1) mode 1: A′(p)=a*A(p−1)+b;
2) mode 2: A′(p)=(a+1)*A(p−1)−a*A(p−2)+b;
3) mode 3: A′(p)=(a-1)*A(p−1)+a*A(p−2)+2b;
4) mode 4: A′(p)=A(p−1)+a*A(p−2)−a*A(p−3)+b;
5) mode 5: A′(p)=(a−1)*A(p−1)+(a−1)*A(p−2)+a*A(p−3)+3b;
6) mode 6: A′(p)=A(p−1)+(a−1)*A(p−2)+a*A(p−3)+a*A(p−4)+3b;
7) mode 7: A′(p)=(a+2)*A(p−1)−(2a+1)*A(p−2)+a*A(p−3)+b.
For the above-disclosed various prediction methods, 1) a predetermined method may be used for a specific unit (e.g., slice, coding block, frame, N units, etc.), and the specific unit used may be transmitted/received through signaling information, or 2) a method of minimizing the error may be transmitted/received by signaling information for every point. When the information is transmitted in the specific unit, an operation of searching for a prediction method that minimizes the total sum of prediction errors per point within the specific unit may be added.
Also, for prediction coefficients a and b, 1) predetermined values may be used and signaling information about the coefficients may be transmitted/received, or 2) signaling information about a method of minimizing the error may be transmitted/received for every point, or 3) the coefficients may be defined as functions inversely proportional to the distance between points.
(3) Performing attribute prediction using a predictive tree structure
As a method for attribute prediction, a predictive tree structure may be used, but a parent different from that in the position prediction may be used. In this case, an operation for separately selecting a parent for attribute prediction is needed, and an optimal attribute prediction method (prediction_type) may be determined: (A) Attribute predictive tree construction and (B) Attribute prediction method.
(A) Attribute predictive tree construction
The parent may be defined through neighbor search within the prediction unit. A neighbor according to the embodiments may be defined as a point having a high similarity to an attribute. For example, a point for which the difference between attributes is minimized may be defined as an attribute neighbor point.
Based on the definition of the attribute neighbor, points in the prediction unit may be sorted in order of the attributes. For example, in the case of reflectance, attribute neighbor search may be performed after sorting points in ascending (or descending) order of reflectance (or a specific order).
In addition, in the case of color, for example, the points may be sorted in ascending (or descending) order of luma (or a specific order), or sorted in ascending (or descending) order (or a specific order) according to the order of green, blue, and red, or colors may be sorted in Morton code order.
After estimating an attribute neighbor relationship for the sorted points, a parent-child relationship may be established to construct a predictive tree as shown in
Terms such as parent, grandparent, etc. according to embodiments may be interpreted and used as a concept continuously referring to an upper node or upper point of a point (node) that is a current target.
For example, in
A relationship between a child point and a parent point may be represented by an arrow pointing from the child point to the parent point.
In this case, some points or vertices 16000 may have a plurality of children.
(B) Attribute prediction method
Attribute prediction according to embodiments may be performed using various attribute prediction methods based on an attribute predictive tree structure. The attribute prediction method may be delivered for each point or according to a specific unit such as a prediction unit: (a) using the attribute prediction mode (Prediction type=0); (b) using the average of neighbor attributes (Prediction_type=1); (c) using the relationship between neighbor attributes (prediction_type=2); (d) delivering a coefficient for the relationship between neighbor attributes (prediction_type=3).
(a) Using the attribute prediction mode (Prediction type=0)
This is a case where a prediction mode is used for a separately constructed predictive tree.
Accordingly, a predictive tree separate from a geometry predictive tree constructed based on positional similarity may be generated. In this case, as proposed above in (2) using a separate attribute prediction mode different from the position prediction mode, the attribute prediction method may be used.
The same tree as used in the geometry coding scheme may be used, and a mode related to use of the tree may be transmitted and received by signaling information.
(b) Using the average of neighbor attributes (Prediction_type=1)
The attribute prediction-based compression method according to the embodiments may perform prediction based on the average of neighbor attributes when similar attributes are sorted in a compression unit. In this case, the number of neighbor attributes to be used for the average of the neighbor attributes may be indicated by num_neighbors, and may be transmitted/received by signaling information.
Neighbors may be signaled based on the actual distances between the points. Starting with the parent, the attribute of a child may be predicted as the average of attributes as many as the number of neighbors according to num_neighbors. For the average, a distance-based weight may be used.
A′(p)=sum[weight(p−k)*A(p−k) ]
(c) Using the relationship between neighbor attributes (prediction_type=2)
According to embodiments, an attribute may be estimated using a tendency of the attribute. The tendency may be predicted through linear regression for the distribution of neighbor attributes, and the prediction coefficients a and b may be delivered directly to the reception device by the transmission device according to the embodiments or may be estimated by the reception device. Here, avg{ A(p)} used for prediction may denote the average of attributes centered on the current point. In addition, the parent index preceding the current point may be delivered directly.
A′(p)=a*avg{A(p)}+b
a=A(p−1)/avg (A(p−1)), b=A(p−1)−a*avg{A(p−2)}
For example, the parent index (Parent_index) indicates an index of a node for obtaining avg {A(p)}. When a and b are not separately delivered, indexes for obtaining avg{A(p−1)} and avg{A(p−2)} may be additionally transmitted.
(d) Delivering a coefficient for the relationship between neighbor attributes (prediction_type=3)
The coefficients may be delivered directly for the tendency of the attribute. In this case, the same coefficient may be applied to as many points as num_same_coeff_pred_points after the current point. By using similar coefficients when points with similar attributes are clustered together in constructing the predictive tree, the bits required to transmit the coefficients may be effectively reduced.
A′(p)=a*A(p−1)+b
For the above-disclosed various prediction methods, 1) a predetermined method may be used in every specific unit (e.g., prediction unit, slice, coding block, frame, N units, etc.), and signaling information about the employed unit may be transmitted/received, or 2) a prediction method may be used for a method of minimizing the error and transmitted/received by signaling information for every point.
When the information is transmitted in the specific unit, an operation of searching for a prediction method that minimizes the total sum of prediction errors per point within the specific unit may be added. Also, for prediction coefficients a and b, 1) predetermined values may be used and signaling information about the prediction coefficients may be transmitted/received, or 2) signaling information about a method of minimizing the error may be transmitted/received for every point, or 3) the prediction coefficients may be defined as functions inversely proportional to the distance between points.
D) Transform
In an environment where data loss is acceptable, duplicate information between residuals may be reduced through transformation. As an example of available transforms, lifting transform may be used.
An attribute signal including attribute data may be split into a high-frequency component H(N) and a low-frequency component L(N), and a difference value D(N) may be generated based on the predicted value P(N) obtained by prediction to generate an update value U(N). The transformation procedure may be performed by adding the update value U(N) and the low-frequency component L(N) and repeating the split, prediction, and update operations again.
In order to reconstruct the attribute data, the difference D(0) may be updated, a predicted value may be generated by performing differentiation from the low-frequency component L′ (0) and added to the difference to generate the high-frequency component H(0), which may be merged with the low-frequency component L(0). The attribute data may be reconstructed by repeating this transformation procedure based on the merged value and the difference D(1).
E) Quantization
Coding efficiency may be increased by delivering a prediction error of the predicted value with respect to the current data.
Res(p)=A(p)−A′(p)
In an environment where data loss is acceptable, the data may further reduced through normalization. In this case, different quantization values may can be applied according to groups. This method may be used in order to prevent an error from being propagated by applying a small normalization value to frequently referenced data.
For example, when reference between groups is allowed, a quantization weight may be applied according to the number of times a group is referenced. Here, the quantization weight Q( )may be defined to be inversely proportional to the number of times the prediction group is referenced.
Quant[x]=FLOOR[Res(p)/Q(referred number of prediction group)]
The quantization weight Q( )may be applied according to the number of times a point is referenced in a predictive coding process according to embodiments. In this regard, when the point has a relationship of a parent, grand parent, and grand-grand parent with neighbor points, and a case where the point is used in the prediction process described in C) is defined to be referenced, the quantization weight may be defined to be inversely proportional to the number of references to the point.
Quant[x]=FLOOR[Res(p)/Q(referred number of vertex)]
The quantization weight Q( )may be applied according to an order in which vertices according to the embodiments are coded from the root. When the number of parent-child relationships from the current vertex to the root (or a vertex serving as a reference) is defined as a vertex distance, the quantization weight may be defined to be proportional to the vertex distance.
Quant[x]=FLOOR[Res(p)/Q(vertex distance from root)]
The quantization weight Q( )may be applied differently according to the number of child nodes according to embodiments. In this case, the quantization weight may be defined to be inversely proportional to the number of children.
Quant[x]=FLOOR[Res(p)/Q(number of children)]
One or more of the quantization weights described above may be used in combination.
The point cloud data transmission method/device according to the embodiments of
In addition, the point cloud data reception method/device according to the embodiments of
The signaling information according to the embodiments may be referred to as metadata, parameters, or the like.
The bitstream according to the embodiments may include encoded point cloud data, and may also include parameters (signaling information, metadata) indicating information required to decode the point cloud data.
The encoded point cloud data in the bitstream may include geometry data and/or attribute data.
The geometry data and/or attribute data may be included in the bitstream on a slice basis.
A geometry slice may include a geometry slice header and geometry slice data.
An attribute slice may include an attribute slice header and attribute slice data.
Information related to compression according to a prediction group according to embodiments may be defined in a parameter set and an attribute slice header (ASH) as follows.
It may be defined in a position corresponding to the application and system or a separate position, such that the scope of application and the method of application may be used differently.
The attribute parameter set (APS) and attribute slice header (ASH) according to the embodiments may include the information related to encoding of the point cloud data according to the embodiments.
Information related to encoding of the point cloud data according to the embodiments may be included in a geometry parameter set (GPS) and a geometry slice header (GSH) so as to be associated with a geometry coding method according to embodiments or applied to geometry coding.
The information related to the encoding of the point cloud data according to the embodiments may also be included in a sequence parameter set (SPS), a tile parameter set (TPS), and the like.
When a syntax element according to the embodiments is applicable to a plurality of point cloud data streams as well as a current point cloud data stream, the information may be carried in a parameter set of a higher concept.
The signaling information according to the embodiments may be generated in the compression (encoding) process of the prediction-based scalable point cloud compression transmission device/method, and may be used in the decoding process of the prediction-based scalable point cloud compression reception device/method.
Hereinafter, parameters (which may be referred to as various terms such as metadata and signaling information) according to embodiments may be generated in a process of a transmitter according to embodiments, which will be described later, and transmitted to a receiver according to the embodiments so as to be used in a reconstruction process. For example, the parameters according to the embodiments may be generated by the metadata processor (or metadata generator) of the transmission device according to the embodiments and acquired by the metadata parser of the reception device according to the embodiments.
Geometry Parameter set Syntax
Attribute parameter set information may be generated, transmitted and received by the point cloud data transmission method/device according to the embodiments of
When pred_attr_enable_flag is equal to 1, it may indicate that prediction-based point cloud attribute compression is used and the related parameters are delivered through the attribute parameter set (aps) or attribute slice or a related parameter container. When equal to 0, it may indicate that prediction-based point cloud attribute compression is not used and another compression method is used.
When independent_pred_mode_enable_flag is equal to 1, it may indicate that a prediction mode separate from the prediction mode used in position compression is used in the prediction-based attribute compression method. When independent_pred_mode_enable_flag is equal to 0, the prediction mode used in the prediction-based position compression may be used in the prediction-based attribute compression.
attr_pred_mode, coeff_a, and coeff_b may signal one mode of prediction methods used in the attribute compression method and related coefficients. When attr_pred_metho, coeff_a, and coeff_b are signaled in the aps, it may be indicated that the same attribute prediction mode (attr_pred_method) is used in the current frame. When they are signaled in a slice header, it may be indicated that the same mode is used in a slice.
Different methods may be signaled according to respective points. Each method may mean the following:
1) mode 1: A′(p)=a*A(p−1)+b;
2) mode 2: A′(p)=(a+1)*A(p−1)−a*A(p−2)+b;
3) mode 3: A′(p)=(a−1)*A(p−1)+a*A(p−2)+2b;
4) mode 4: A′(p)=A(p−1)+a*A(p−2)−a*A(p−3)+b;
5) mode 5: A′(p)=(a−1)*A(p−1)+(a−1)*A(p−2)+a*A(p−3)+3b;
6) mode 6: A′(p)=A(p−1)+(a−1)*A(p−2)+a*A(p−3)+a*A(p−4)+3b;
7) mode 7: A′(p)=(a+2)*A(p−1)−(2a+1)*A(p−2)+a*A(p−3)+b.
When independent_parent_enable_flag is equal to 1, it may indicate that a tree structure separate from the tree structure used in position compression is used in the prediction-based attribute compression method. When is equal to 0, it indicates that the tree structure used in prediction-based position compression is used in prediction-based attribute compression.
Prediction_type may indicate an attribute prediction method. Among the values thereof, 0 may indicate a method of delivering the attribute prediction modep; 1 may indicate a prediction method based on the average of neighbor points; 2 may indicate a prediction method based on the tendency of an attribute based on multiple parents; and 3 may indicate prediction using coefficients based on a linear relationship with the parent.
num_neighbors may indicate the number of neighbor points, and may indicate the number of points from the parent of a point to be predicted toward the root according to embodiments.
Num_attr_parents_minus1 plus 1 may indicate the number of parents used for attribute prediction.
correlated_data_search_method indicates a method for searching for the parent of each point. The values thereof may be defined as follows: 0 may indicate input order; 1 may indicate time stamp order; 2 may indicate Morton code order; 3 may indicate an order according to a radius from the center axis of LiDAR; 4 may indicate an order according to the horizontal angle from the center axis of LiDAR; 5 may indicate an order according to the vertical angle of the LiDAR.
transform_enable_flag equal to 1 may indicate that a transform according to embodiments is used.
qnant_mode may indicate a method of determining a quantization weight. Among the values thereof, 0 may indicate that quantization is not used; 1 may indicate that a weight is determined according to the number of times a group is referenced; 2 may indicate that the weight is determined according to the number of times a vertex is referenced; 3 may indicate that the weight is determined according to an order in which the vertices are coded from the root; 4 may indicate that the weight is determined according to the number of child nodes.
When pred_attr_enable_flag is 1, independent_parent_enable_flag may be provided.
When independent_parent_enable_flag is not 1, attr_pred_mode, coeff_a, and coeff_b may be provided.
When independent_parent_enable_flag is 1, specific information according to prediction_type and type may be delivered as follows. According to the prediction type/mode according to the embodiments, the method/device for receiving point cloud data according to the embodiments may reconstruct the point cloud data based on attr_pred_mode, num_neighbors, num_attr_parents_minus1, and parent_index[i] according to num_attr_parents_minus1, and the like.
The method/device according to the embodiments may reconstruct the point cloud data based on correlated_data_search_method, transform_enable_flag[i], quant_mode[i], and the like.
by the point cloud data transmission method/device according to the embodiments of
The bitstream according to the embodiments includes an attribute slice header and attribute slice data.
The definition of the elements described in
parent_index may indicate a parent index used in geometry and attribute predictions according to embodiments. According to embodiments, when the attribute information is transmitted according to the parent-child order, the bitrate may be reduced by skipping separate transmission of parent_index.
num_same_coeff_pred_points may indicate the number of points predicted using the same coefficients.
coeff_c1 and coeff_c2 indicate coefficients used for prediction using the same coefficients.
Residual indicates a prediction error.
According to the prediction type according to the embodiments, point cloud data may be reconstructed based on num_neighbors, num_attr_parents_minus1, num_same_coeff pred_points[i], coeff_c1, coeff_c2, and the like.
The attribute slice data according to the embodiments may include an attribute dimension (AttrDim).
children_count[i] indicating the number of child points as many as numPoints may be provided.
When independent_parent_enable_flag is not 1, attr_pred_mode[i], coeff_a[i], and coeff_b[i] may be transmitted for every i-th point.
When independent_parent_enable_flag is 1, parent_index[i] and prediction_type[i] may be delivered. Also, attr_pred_mode[i], num_neighbors[i], and num_attr_parents_minus1[i] according to prediction_type[i] may be delivered.
And parent_index [i][j], num_same_coeff pred_points[i], coeff_cl[i], coeff_c2[i] according to num_attr_parents_minus1[i].
residual[i][j] may be delivered for each AttrDim.
For example, when prediction type=3, signal information may be transmitted in slice data as shown in
Furthermore, according to embodiments, num_same_coeff_pred_points, coeff_ c1, and coeffc2 according to prediction type=3 may be carried in the attribute parameter set as shown in
A point cloud data transmission method according to embodiments may include encoding point cloud data and transmitting the point cloud data.
The encoding of the point cloud data according to the embodiments may include generating the above-described predictive tree based on geometry data of the point cloud data, and predicting the geometry data and attribute data of the point cloud data based on the predictive tree.
The encoding of the point cloud data according to the embodiments may include generating a bitstream containing the point cloud data and parameter information related to the point cloud data, wherein the bitstream may further contain coefficient information and mode information related to prediction of the attribute data.
The encoding of the point cloud data according to the embodiments may further include generating predictive data for the attribute data based on an error between the attribute data of the point cloud data.
The encoding of the point cloud data according to embodiments may further include generating an attribute predictive tree based on similarity between attribute data of the point cloud data.
The encoding of the point cloud data according to the embodiments may include generating an attribute predictive tree based on similarity between attribute data of the point cloud data, and generating predictive data for the attribute data based on the attribute predictive tree and an average of neighbor attribute data of the attribute data.
The encoding of the point cloud data according to the embodiments may include generating an attribute predictive tree based on similarity between attribute data of the point cloud data, and generating predictive data for the attribute data based on the attribute predictive tree and one or more neighbor attribute data of the attribute data.
The encoding of the point cloud data according to the embodiments may include generating an attribute predictive tree based on similarity between attribute data of the point cloud data, and generating predictive data for the attribute data based on the attribute predictive tree, one or more neighbor attribute data of the attribute data, and coefficient information related to the one or more neighbor attribute data.
The encoding of the point cloud data according to the embodiments may include generating a bitstream containing the point cloud data and parameter information related to the point cloud data, wherein the bitstream may contain parameter information related to prediction of attribute data.
A device for receiving point cloud data according to embodiments may include a memory and a processor connected to the memory, wherein the processor may receive point cloud data and decode the point cloud data.
The processor according to the embodiments may generate a predictive tree based on the geometry data of the point cloud data, and predict the geometry data and attribute data of the point cloud data based on the predictive tree.
The processor according to the embodiments may receive a bitstream containing the point cloud data and parameter information related to the point cloud data, wherein the bitstream may contain coefficient information and mode information related to prediction of the attribute data.
The processor according to the embodiments may generate predictive data for attribute data of the point cloud data based on an error between the attribute data.
The processor according to the embodiments may generate an attribute predictive tree based on similarity between attribute data of the point cloud data.
The processor according to the embodiments may generate an attribute predictive tree based on similarity between attribute data of the point cloud data.
It may generate predictive data for the attribute data based on the attribute predictive tree and an average of neighbor attribute data of the attribute data.
The processor according to the embodiments may generate an attribute predictive tree based on similarity between attribute data of the point cloud data, and generate predictive data for the attribute data based on the attribute predictive tree and one or more neighbor attribute data of the attribute data.
The processor according to the embodiments may generate an attribute predictive tree based on similarity between attribute data of the point cloud data, and generate predictive data for the attribute data based on the attribute predictive tree, one or more neighbor attribute data of the attribute data, and coefficient information related to the one or more attribute data.
The processor according to the embodiments may receive a bitstream containing the point cloud data and parameter information related to the point cloud data, wherein the bitstream may contain parameter information related to prediction of the attribute data.
The operations of the above-described embodiments may be performed through the components of the prediction-based point cloud compression transmission/reception device/method according to the embodiments described below. Each component according to the embodiments may correspond to hardware, software, a processor, and/or a combination thereof.
The prediction-based point cloud attribute compression method according to the embodiments may be implemented as follows. A method of compressing position information (geometry) about the point cloud data may be applied to position information compression and other compression methods in the same/similar manner.
The configuration and/or operation of a point cloud encoder, for example, an attribute encoder, of the transmission device according to the embodiments may be as follows.
Regarding predictive tree generation (the predictive tree generation in
In this case, when independent_pred_mode_enable_flag is 0, the operation of generating a predictive tree structure and selecting a prediction mode may be skipped, and the tree structure and prediction method used for position information compression may be used.
When Independent_pred_mode_enable_flag is 1 and independent_parent_enable_flag is 0, the operation of generating a predictive tree structure may be skipped. A tree structure used for position information compression may be used, and prediction mode selection may be performed based on the tree structure.
When independent_pred_mode_enable_flag is 1 and independent_parent_enable_flag is 1, different prediction methods may be used according to prediction_type.
After performing quantization and transform (the transform and quantization in
In this case, when quantization occurs, errors that may occur in receiver may be reduced by updating the quantized value during prediction. The execution order of the blocks in
The configuration and/or operation of a point cloud decoder, for example, an attribute decoder, of the reception device according to the embodiments may be as follows.
The reception method/device according to the embodiments reconstructs a signal (point cloud data) in reverse of the execution order of the transmission method/device according to the embodiments.
Entropy decoding is performed on the received bitstream.
After performing inverse quantization and inverse transform, reconstruction is performed.
In this regard, the point cloud data reception method/device according to the embodiments may identify the tree structure based on the parent index transmitted from the transmission method/device (see
Regarding the prediction and reconstruction according to the embodiments, the reception method/device according to the embodiments may predict point cloud data according to a prediction method carried out by the transmitter to generate predictive data, and reconstruct the point cloud data based on the predictive data and residual. The point cloud data may be reconstructed by summing the predictive data and the residual.
In this case, the receiver does not perform the prediction group information generation, and therefore the execution time may be greatly reduced compared to that of the transmitter. In addition, when the prediction operation can be skipped, the execution time may be further reduced.
When Independent_pred_mode_enable_flag is set to 0, attribute coding is performed based on the position predictive tree and the position prediction mode.
When Independent_pred_mode_enable_flag is set to 1 and independent_parent_enable_flag is set to 0, the position predictive tree is used, but the attribute prediction mode used is different from the position prediction mode.
When independent_parent_enable_flag is set to 0, attr_pred_mode, coeff_a, and coeff_b are transmitted by signaling information as shown in
When Independent_parent_enable_flag is set to 1, an attribute predictive tree according to embodiments is constructed and an attribute prediction mode is obtained.
When the predictive tree is configured, parent_index may be signaled (parent_index[i] in
After prediction mode selection, the difference between the source data and the predicted data is obtained and encoded through transform, quantization, and entropy coding.
A bitstream containing the encoded point cloud data and/or signaling information may be transmitted.
In the predictive tree generation, an attribute predictive tree representing a parent-child node relationship as described with reference to
In the prediction mode selection, attribute data may be attribute-coded based on prediction modes according to prediction types 0 to 3 described above.
In the residual generation, a residual may be generated based on a difference between the predictive data generated through prediction and the attribute data.
In the transform & quantization, lifting transform and quantization may be applied to transmit the residual more efficiently.
In the entropy coding, the transformed and/or quantized residual bitstream may be entropy-encoded.
The point cloud data reception method/device according to the embodiments reconstructs a residual, which is a predicted error for each point, through entropy decoding, inverse quantization, and inverse transformation of the received bitstream.
The parent of the current point is searched for based on quantization-related grouping information (e.g. qnant_mode) and a correlated data search method included in the received bitstream.
A predicted value is searched for using a parent and related points based on information about a prediction method (the syntax of
The point is reconstructed based on the received residual (e.g. residual[i][j] in
Regarding the prediction and reconstruction according to the embodiments, the reception method/device according to the embodiments may predict point cloud data according to a prediction method carried out by the transmitter to generate predictive data, and reconstruct the point cloud data based on the predictive data and residual. The point cloud data may be reconstructed by summing the predictive data and the residual.
The reception method/device according to the embodiments may decode attribute data based on pred_attr_enable_flag.
When independent_pred_mode_enable_flag is TRUE and independent_parent_enable_flag is not TRUE, predictive decoding of the attribute data may be performed based on attr_pred_mode, coeff_a, and coeff_b.
When independent_pred_mode_enable_flag is TRUE and independent_parent_enable_flag is TRUE, predictive decoding may be performed according to the predictive coding according to prediction_type. According to a value indicated by prediction_type, the attribute data may be predicted and decoded based on attr_pred_mod, num_neighbors, num_attr_parents_minus1, parent_index[i], num_same_coeff_pred_points[i], coeff_c1[i], coeff_c2[i], and the like.
When independent_pred_mode_enable_flag is not TRUE, attribute predictive decoding may be performed in the same way as the position prediction.
In the entropy decoding, entropy decoding may be performed on a bitstream containing point cloud data.
In the inverse quantization & transformation, the point cloud data is inversely quantized and inversely transformed in the reverse of the transmission operation.
In the correlated data search, an upper point (parent point) for decoding the current point is searched for. The correlated data of the current point may be searched for by a method according to correlated_data_search_method.
In the prediction, attribute predictive data for the current point is generated based on the parameter information of
In the reconstruction, the point cloud data may be reconstructed by summing the prediction data generated through the attribute prediction and the received residual.
S24000: A method for transmitting point cloud data according to embodiments includes encoding point cloud data.
The encoding operation according to the embodiments may include operations of the transmission device 10000 and the point cloud video encoder 10002 in
S24010: The point cloud data transmission method according to the embodiments further includes transmitting the point cloud data.
The transmission operation according to the embodiments may include the operations of the transmitter 10003 of the transmission device 10000 in
S25000: A method for receiving point cloud data according to embodiments includes receiving point cloud data.
The reception operation according to the embodiments may include the operations of the reception device 10004 and the point cloud video receiver 10005 in
S25010: The point cloud data reception method according to the embodiments further includes decoding the point cloud data.
The decoding operation according to the embodiments may include the operations of the point cloud video decoder 10006 in
The methods/devices according to the embodiments may provide high-speed decoding and encoding for an environment requiring low delay by using a prediction-based point cloud compression method.
In addition, by using a prediction unit group, only required unit information may be transmitted or received. Thereby, decoding may be adaptively performed in a transmission/reception environment, and a coding method suitable for a low delay environment may be provided.
The embodiments have been described in terms of a method and/or a device. The description of the method and the description of the device may complement each other.
Although embodiments have been described with reference to each of the accompanying drawings for simplicity, it is possible to design new embodiments by merging the embodiments illustrated in the accompanying drawings. If a recording medium readable by a computer, in which programs for executing the embodiments mentioned in the foregoing description are recorded, is designed by those skilled in the art, it may also fall within the scope of the appended claims and their equivalents. The devices and methods may not be limited by the configurations and methods of the embodiments described above. The embodiments described above may be configured by being selectively combined with one another entirely or in part to enable various modifications. Although preferred embodiments have been described with reference to the drawings, those skilled in the art will appreciate that various modifications and variations may be made in the embodiments without departing from the spirit or scope of the disclosure described in the appended claims. Such modifications are not to be understood individually from the technical idea or perspective of the embodiments.
Various elements of the devices of the embodiments may be implemented by hardware, software, firmware, or a combination thereof. Various elements in the embodiments may be implemented by a single chip, for example, a single hardware circuit. According to embodiments, the components according to the embodiments may be implemented as separate chips, respectively. According to embodiments, at least one or more of the components of the device according to the embodiments may include one or more processors capable of executing one or more programs. The one or more programs may perform any one or more of the operations/methods according to the embodiments or include instructions for performing the same. Executable instructions for performing the method/operations of the device according to the embodiments may be stored in a non-transitory CRM or other computer program products configured to be executed by one or more processors, or may be stored in a transitory CRM or other computer program products configured to be executed by one or more processors. In addition, the memory according to the embodiments may be used as a concept covering not only volatile memories (e.g., RAM) but also nonvolatile memories, flash memories, and PROMs. In addition, it may also be implemented in the form of a carrier wave, such as transmission over the Internet. In addition, the processor-readable recording medium may be distributed to computer systems connected over a network such that the processor-readable code may be stored and executed in a distributed fashion.
In this document, the term “/” and “,” should be interpreted as indicating “and/or.” For instance, the expression “A/B” may mean “A and/or B.” Further, “A, B” may mean “A and/or B.” Further, “A/B/C” may mean “at least one of A, B, and/or C.” “A, B, C” may also mean “at least one of A, B, and/or C.” Further, in the document, the term “or” should be interpreted as “and/or.” For instance, the expression “A or B” may mean 1) only A, 2) only B, and/or 3) both A and B. In other words, the term “or” in this document should be interpreted as “additionally or alternatively.”
Terms such as first and second may be used to describe various elements of the embodiments. However, various components according to the embodiments should not be limited by the above terms. These terms are only used to distinguish one element from another. For example, a first user input signal may be referred to as a second user input signal. Similarly, the second user input signal may be referred to as a first user input signal. Use of these terms should be construed as not departing from the scope of the various embodiments. The first user input signal and the second user input signal are both user input signals, but do not mean the same user input signal unless context clearly dictates otherwise.
The terminology used to describe the embodiments is used for the purpose of describing particular embodiments only and is not intended to be limiting of the embodiments. As used in the description of the embodiments and in the claims, the singular forms “a”, “an”, and “the” include plural referents unless the context clearly dictates otherwise. The expression “and/or” is used to include all possible combinations of terms. The terms such as “includes” or “has” are intended to indicate existence of figures, numbers, steps, elements, and/or components and should be understood as not precluding possibility of existence of additional existence of figures, numbers, steps, elements, and/or components. As used herein, conditional expressions such as “if” and “when” are not limited to an optional case and are intended to be interpreted, when a specific condition is satisfied, to perform the related operation or interpret the related definition according to the specific condition.
Operations according to the embodiments described in this specification may be performed by a transmission/reception device including a memory and/or a processor according to embodiments. The memory may store programs for processing/controlling the operations according to the embodiments, and the processor may control various operations described in this specification. The processor may be referred to as a controller or the like. In embodiments, operations may be performed by firmware, software, and/or combinations thereof. The firmware, software, and/or combinations thereof may be stored in the processor or the memory.
The operations according to the above-described embodiments may be performed by the transmission device and/or the reception device according to the embodiments. The transmission/reception device may include a transmitter/receiver configured to transmit and receive media data, a memory configured to store instructions (program code, algorithms, flowcharts and/or data) for the processes according to the embodiments, and a processor configured to control the operations of the transmission/reception device.
The processor may be referred to as a controller or the like, and may correspond to, for example, hardware, software, and/or a combination thereof. The operations according to the above-described embodiments may be performed by the processor. In addition, the processor may be implemented as an encoder/decoder for the operations of the above-described embodiments.
As described above, related details have been described in the best mode for carrying out the embodiments.
As described above, the embodiments are fully or partially applicable to a point cloud transmission/reception device and system.
Those skilled in the art may change or modify the embodiments in various ways within the scope of the embodiments.
Embodiments may include variations/modifications within the scope of the claims and their equivalents.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10-2020-0050274 | Apr 2020 | KR | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/KR2021/004983 | 4/21/2021 | WO |
