Apple Patent | Geometry information signaling for occluded points in an occupancy map video

Patent: Geometry information signaling for occluded points in an occupancy map video

Drawings: Click to check drawins

Publication Number: 20210211703

Publication Date: 20210708

Applicant: Apple

Abstract

In an example method, points that represent three-dimensional visual volumetric content are received, and patches are determined, where each patch corresponds to a respective portion of the visual volumetric content. A patch image representing a set of points corresponding to the patch projected onto a respective patch plane is generated for each patch. The patch images are packed into image frames, and the image frames are encoded. An occupancy map corresponding to the image frames is generated. The occupancy map indicates, for each image frame: locations of the patch images in the image frame, and depth information of sets of points corresponding to the patch images in the image frame. The depth information indicates, for each patch image, depths of the set of points corresponding to the patch image in a direction perpendicular to a patch plane of the patch image.

Claims

1. A device comprising: one or more processors; and memory storing instructions that when executed by the one or more processors, cause the one or more processors to perform operations comprising: receiving a plurality of points that represent three-dimensional visual volumetric content; determining, for the three-dimensional visual volumetric content, a plurality of patches, wherein each patch corresponds to a respective portion of the three-dimensional visual volumetric content; generating, for each patch, a patch image representing a set of points corresponding to the patch projected onto a respective patch plane; packing the patch images into one or more image frames; encoding the one or more image frames; and generating an occupancy map corresponding to the one or more image frames, wherein the occupancy map indicates, for each image frame: locations of one or more of the patch images in the image frame, and depth information of one or more sets of points corresponding to the one or more of the patch images in the image frame, wherein the depth information indicates, for each patch image, depths of the set of points corresponding to the patch image in a direction perpendicular to a patch plane of the patch image.

2. The device of claim 1, wherein the occupancy map comprises, for each patch image, a respective plurality of first elements, wherein each first element corresponds to a respective point on the patch plane of the patch image, and wherein each first element indicates respective depths of the points of the set of points corresponding to the patch image along a respective projection line, the projection line extending from the respective point on the patch plane in the direction perpendicular to the patch plane.

3. The device of claim 2, wherein each first element is determined based on a determination whether the set of points corresponding to the patch image comprises any points along the respective projection line.

4. The device of claim 2, wherein each first element is determined based on the depth of each point of the set of points corresponding to the patch image along the respective projection line.

5. The device of claim 2, wherein each first element comprises a respective encoded value indicating the depth of each point of the set of points corresponding to the patch image along the respective projection line.

6. The device of claim 5, wherein the encoded value is determined based on a binary representation of the depths of at least some of the points of the set of points corresponding to the patch image along the respective projection line.

7. The device of claim 2, the operations further comprising down-sampling a spatial resolution of the occupancy map relative to a spatial resolution of the one or more image frames.

8. The device of claim 7, wherein down-sampling the spatial resolution of the occupancy map comprises: determining a plurality of second elements based on the first elements, wherein each second element represents two or more respective first elements.

9. The device of claim 8, wherein determining each second element comprises: identifying two or more respective first elements; comparing, with respect to the two or more respective first elements, the depths of the points of the set of points corresponding to the patch image along the respective projection lines, and determining the second element based on the comparison.

10. The device of claim 8, wherein the comparison comprises a bitwise binary operation.

11. The device of claim 8, wherein the bitwise binary operation comprises a bitwise OR operation or a bitwise AND operation.

12. The device of claim 1, wherein each image frame comprises a respective attribute image portion, wherein the attribute image portion is separated spatially from the patch images in the image frame, and wherein the attribute image portion indicates additional attribute information regarding at least one of the patch images in the image frame.

13. The device of claim 12, wherein the attribute image portion comprises a plurality of attribute image sub-portions, each attribute image sub-portion indicating respective additional attribute information regarding a respective patch image in the image frame.

14. The device of claim 12, wherein each of the attribute image sub-portions are equal in size spatially.

15. The device of claim 12, wherein each attribute image sub-portion comprises: an indication of a location of the attribute image sub-portion in the image frame, and a spatial size of the attribute image sub-portion.

16. The device of claim 15, wherein each attribute image sub-portion comprises: an indication of a patch image in the image frame corresponding to the attribute image sub-portion.

17. The device of claim 15, wherein each attribute image sub-portion comprises: an indication of multiple patch images in the image frame corresponding to the attribute image sub-portion.

18. The device of claim 1, wherein the one or more image frames are encoded in accordance with the high efficiency video coding (HEVC) standard.

19. The device of claim 1, wherein each point comprises spatial information regarding the point and attribute information regarding the point.

20. A method comprising: receiving a plurality of points that represent three-dimensional visual volumetric content; determining, for the three-dimensional visual volumetric content, a plurality of patches, wherein each patch corresponds to a respective portion of the three-dimensional visual volumetric content; generating, for each patch, a patch image representing a set of points corresponding to the patch projected onto a respective patch plane; packing the patch images into one or more image frames; encoding the one or more image frames; and generating an occupancy map corresponding to the one or more image frames, wherein the occupancy map indicates, for each image frame: locations of one or more of the patch images in the image frame, and depth information of one or more sets of points corresponding to the one or more of the patch images in the image frame, wherein the depth information indicates, for each patch image, depths of the set of points corresponding to the patch image in a direction perpendicular to a patch plane of the patch image.

21. A non-transitory, computer-readable storage medium having instructions stored thereon, that when executed by one or more processors, cause the one or more processors to perform operations comprising: receiving a plurality of points that represent three-dimensional visual volumetric content; determining, for the three-dimensional visual volumetric content, a plurality of patches, wherein each patch corresponds to a respective portion of the three-dimensional visual volumetric content; generating, for each patch, a patch image representing a set of points corresponding to the patch projected onto a respective patch plane; packing the patch images into one or more image frames; encoding the one or more image frames; and generating an occupancy map corresponding to the one or more image frames, wherein the occupancy map indicates, for each image frame: locations of one or more of the patch images in the image frame, and depth information of one or more sets of points corresponding to the one or more of the patch images in the image frame, wherein the depth information indicates, for each patch image, depths of the set of points corresponding to the patch image in a direction perpendicular to a patch plane of the patch image.

Description

CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority from U.S. Provisional Application Ser. No. 62/958,229, filed on Jan. 7, 2020, which is incorporated herein by reference in its entirety.

TECHNICAL FIELD

[0002] This disclosure relates generally to compression and decompression of point clouds including a plurality of points, each having associated spatial information and attribute information.

BACKGROUND

[0003] Various types of sensors, such as light detection and ranging (LIDAR) systems, 3-D-cameras, 3-D scanners, etc. may capture data indicating positions of points in three-dimensional space, for example positions in the X, Y, and Z planes. Also, such systems may further capture attribute information in addition to spatial information for the respective points, such as color information (e.g., RGB values), texture information, intensity attributes, reflectivity attributes, motion related attributes, modality attributes, or various other attributes. In some circumstances, additional attributes may be assigned to the respective points, such as a time-stamp when the point was captured. Points captured by such sensors may make up a “point cloud” including a set of points each having associated spatial information and one or more associated attributes. In some circumstances, a point cloud may include thousands of points, hundreds of thousands of points, millions of points, or even more points. Also, in some circumstances, point clouds may be generated, for example in software, as opposed to being captured by one or more sensors. In either case, such point clouds may include large amounts of data and may be costly and time-consuming to store and transmit.

SUMMARY

[0004] In an aspect, a method includes receiving a plurality of points that represent three-dimensional visual volumetric content; determining, for the three-dimensional visual volumetric content, a plurality of patches, where each patch corresponds to a respective portion of the three-dimensional visual volumetric content; generating, for each patch, a patch image representing a set of points corresponding to the patch projected onto a respective patch plane; packing the patch images into one or more image frames; encoding the one or more image frames; and generating an occupancy map corresponding to the one or more image frames. The occupancy map indicates, for each image frame: locations of one or more of the patch images in the image frame, and depth information of one or more sets of points corresponding to the one or more of the patch images in the image frame. The depth information indicates, for each patch image, depths of the set of points corresponding to the patch image in a direction perpendicular to a patch plane of the patch image.

[0005] Implementations of this aspect can include one or more of the following features.

[0006] In some implementations, the occupancy map can include, for each patch image, a respective plurality of first elements. Each first element can correspond to a respective point on the patch plane of the patch image. Each first element can indicate respective depths of the points of the set of points corresponding to the patch image along a respective projection line, the projection line extending from the respective point on the patch plane in the direction perpendicular to the patch plane.

[0007] In some implementations, each first element can be determined based on a determination whether the set of points corresponding to the patch image includes any points along the respective projection line.

[0008] In some implementations, each first element can be determined based on the depth of each point of the set of points corresponding to the patch image along the respective projection line.

[0009] In some implementations, each first element can include a respective encoded value indicating the depth of each point of the set of points corresponding to the patch image along the respective projection line.

[0010] In some implementations, the encoded value can be determined based on a binary representation of the depths of at least some of the points of the set of points corresponding to the patch image along the respective projection line.

[0011] In some implementations, the method can further include down-sampling a spatial resolution of the occupancy map relative to a spatial resolution of the one or more image frames.

[0012] In some implementations, down-sampling the spatial resolution of the occupancy map can include determining a plurality of second elements based on the first elements, where each second element represents two or more respective first elements.

[0013] In some implementations, determining each second element can include identifying two or more respective first elements; comparing, with respect to the two or more respective first elements, the depths of the points of the set of points corresponding to the patch image along the respective projection lines, and determining the second element based on the comparison.

[0014] In some implementations, the comparison can include a bitwise binary operation.

[0015] In some implementations, the bitwise binary operation can include a bitwise OR operation or a bitwise AND operation.

[0016] In some implementations, each image frame can include a respective attribute image portion, where the attribute image portion is separated spatially from the patch images in the image frame, and where the attribute image portion indicates additional attribute information regarding at least one of the patch images in the image frame.

[0017] In some implementations, the attribute image portion can include a plurality of attribute image sub-portions, each attribute image sub-portion indicating respective additional attribute information regarding a respective patch image in the image frame.

[0018] In some implementations, each of the attribute image sub-portions can be equal in size spatially.

[0019] In some implementations, each attribute image sub-portion can include an indication of a location of the attribute image sub-portion in the image frame, and a spatial size of the attribute image sub-portion.

[0020] In some implementations, each attribute image sub-portion can include an indication of a patch image in the image frame corresponding to the attribute image sub-portion.

[0021] In some implementations, each attribute image sub-portion can include an indication of multiple patch images in the image frame corresponding to the attribute image sub-portion.

[0022] In some implementations, the one or more image frames can be encoded in accordance with the high efficiency video coding (HEVC) standard or some other image or video coding standard or specification.

[0023] In some implementations, each point can include spatial information regarding the point and attribute information regarding the point.

[0024] Other implementations are directed to systems, devices, and non-transitory, computer-readable media having instructions stored thereon, that when executed by one or more processors, cause the one or more processors to perform operations described herein.

[0025] The details of one or more embodiments are set forth in the accompanying drawings and the description below. Other features and advantages will be apparent from the description and drawings, and from the claims.

BRIEF DESCRIPTION OF DRAWINGS

[0026] FIG. 1 illustrates a system including a sensor that captures information for points of a point cloud and an encoder that compresses spatial information and attribute information of the point cloud, where the compressed spatial and attribute information is sent to a decoder.

[0027] FIG. 2A illustrates components of an encoder for encoding intra point cloud frames.

[0028] FIG. 2B illustrates components of a decoder for decoding intra point cloud frames.

[0029] FIG. 3A illustrates an example patch segmentation process.

[0030] FIG. 3B illustrates an example image frame including packed patch images and padded portions.

[0031] FIG. 3C illustrates an example image frame including overlapping patches.

[0032] FIG. 3D illustrates a point cloud being projected onto multiple projections.

[0033] FIG. 3E illustrates a point cloud being projected onto multiple parallel projections.

[0034] FIG. 4 illustrates an example process of generating geometry and occupancy maps representing one or more points in a point cloud.

[0035] FIG. 5 illustrates another example process of generating geometry and occupancy maps representing one or more points in a point cloud.

[0036] FIG. 6 illustrates example schemes for down-sampling an occupancy map.

[0037] FIG. 7 illustrates additional example schemes for down-sampling an occupancy map when the occupancy map has the depth information.

[0038] FIG. 8A illustrates an example scheme for a threshold based non-binary occupancy map.

[0039] FIG. 8B illustrates an example segmentation of an occupancy range.

[0040] FIG. 9 illustrates an example scheme for generating a multi-threshold non-binary occupancy map.

[0041] FIG. 10 shows an image frame including an example occupancy map, and an image frame including a corresponding attribute map.

[0042] FIG. 11 illustrates an example process for generating information regarding a point cloud.

[0043] FIG. 12 illustrates an example process for using compressed point cloud information in a 3-D telepresence application.

[0044] FIG. 13 illustrates an example process for using compressed point cloud information in a virtual reality application.

[0045] FIG. 14 illustrates an example computer system that may implement an encoder or decoder.

[0046] This specification includes references to “one embodiment” or “an embodiment.” The appearances of the phrases “in one embodiment” or “in an embodiment” do not necessarily refer to the same embodiment. Particular features, structures, or characteristics may be combined in any suitable manner consistent with this disclosure.

[0047] “Comprising.” This term is open-ended. As used in the appended claims, this term does not foreclose additional structure or steps. Consider a claim that recites: “An apparatus comprising one or more processor units … .” Such a claim does not foreclose the apparatus from including additional components (e.g., a network interface unit, graphics circuitry, etc.).

[0048] “Configured To.” Various units, circuits, or other components may be described or claimed as “configured to” perform a task or tasks. In such contexts, “configured to” is used to connote structure by indicating that the units/circuits/components include structure (e.g., circuitry) that performs those task or tasks during operation. As such, the unit/circuit/component can be said to be configured to perform the task even when the specified unit/circuit/component is not currently operational (e.g., is not on). The units/circuits/components used with the “configured to” language include hardware–for example, circuits, memory storing program instructions executable to implement the operation, etc. Reciting that a unit/circuit/component is “configured to” perform one or more tasks is expressly intended not to invoke 35 U.S.C. .sctn. 112(f), for that unit/circuit/component. Additionally, “configured to” can include generic structure (e.g., generic circuitry) that is manipulated by software and/or firmware (e.g., an FPGA or a general-purpose processor executing software) to operate in manner that is capable of performing the task(s) at issue. “Configure to” may also include adapting a manufacturing process (e.g., a semiconductor fabrication facility) to fabricate devices (e.g., integrated circuits) that are adapted to implement or perform one or more tasks.

[0049] “First,” “Second,” etc. As used herein, these terms are used as labels for nouns that they precede, and do not imply any type of ordering (e.g., spatial, temporal, logical, etc.). For example, a buffer circuit may be described herein as performing write operations for “first” and “second” values. The terms “first” and “second” do not necessarily imply that the first value must be written before the second value.

[0050] “Based On.” As used herein, this term is used to describe one or more factors that affect a determination. This term does not foreclose additional factors that may affect a determination. That is, a determination may be solely based on those factors or based, at least in part, on those factors. Consider the phrase “determine A based on B.” While in this case, B is a factor that affects the determination of A, such a phrase does not foreclose the determination of A from also being based on C. In other instances, A may be determined based solely on B.

DETAILED DESCRIPTION

[0051] As data acquisition and display technologies have become more advanced, the ability to capture point clouds including thousands or millions of points in 2-D or 3-D space, such as via LIDAR systems, has increased. Also, the development of advanced display technologies, such as virtual reality or augmented reality systems, has increased potential uses for point clouds. However, point cloud files are often very large and may be costly and time-consuming to store and transmit. For example, communication of point clouds over private or public networks, such as the Internet, may require considerable amounts of time and/or network resources, such that some uses of point cloud data, such as real-time uses, may be limited. Also, storage requirements of point cloud files may consume a significant amount of storage capacity of devices storing the point cloud files, which may also limit potential applications for using point cloud data.

[0052] In some embodiments, an encoder may be used to generate a compressed point cloud to reduce costs and time associated with storing and transmitting large point cloud files. In some embodiments, a system may include an encoder that compresses attribute and/or spatial information of a point cloud file such that the point cloud file may be stored and transmitted more quickly than non-compressed point clouds and in a manner that the point cloud file may occupy less storage space than non-compressed point clouds.

[0053] In some embodiments, compression of attributes of points in a point cloud may enable a point cloud to be communicated over a network in real-time or in near real-time. For example, a system may include a sensor that captures attribute information about points in an environment where the sensor is located, where the captured points and corresponding attributes make up a point cloud. The system may also include an encoder that compresses the captured point cloud attribute information. The compressed attribute information of the point cloud may be sent over a network in real-time or near real-time to a decoder that decompresses the compressed attribute information of the point cloud. The decompressed point cloud may be further processed, for example to make a control decision based on the surrounding environment at the location of the sensor. The control decision may then be communicated back to a device at or near the location of the sensor, where the device receiving the control decision implements the control decision in real-time or near real-time. In some embodiments, the decoder may be associated with an augmented reality system and the decompressed attribute information may be displayed or otherwise used by the augmented reality system. In some embodiments, compressed attribute information for a point cloud may be sent with compressed spatial information for points of the point cloud. In other embodiments, spatial information and attribute information may be separately encoded and/or separately transmitted to a decoder.

[0054] In some embodiments, a system may include a decoder that receives one or more sets of point cloud data including compressed attribute information via a network from a remote server or other storage device that stores the one or more point cloud files. For example, a 3-D display, a holographic display, or a head-mounted display may be manipulated in real-time or near real-time to show different portions of a virtual world represented by point clouds. In order to update the 3-D display, the holographic display, or the head-mounted display, a system associated with the decoder may request point cloud data from the remote server based on user manipulations of the displays, and the point cloud data may be transmitted from the remote server to the decoder and decoded by the decoder in real-time or near real-time. The displays may then be updated with updated point cloud data responsive to the user manipulations, such as updated point attributes.

[0055] In some embodiments, a system, may include one or more LIDAR systems, 3-D cameras, 3-D scanners, etc., and such sensor devices may capture spatial information, such as X, Y, and Z coordinates for points in a view of the sensor devices. In some embodiments, the spatial information may be relative to a local coordinate system or may be relative to a global coordinate system (e.g., a Cartesian coordinate system may have a fixed reference point, such as a fixed point on the earth, or may have a non-fixed local reference point, such as a sensor location).

[0056] In some embodiments, such sensors may also capture attribute information for one or more points, such as color attributes, reflectivity attributes, velocity attributes, acceleration attributes, time attributes, modalities, and/or various other attributes. In some embodiments, other sensors, in addition to LIDAR systems, 3-D cameras, 3-D scanners, etc., may capture attribute information to be included in a point cloud. For example, in some embodiments, a gyroscope or accelerometer, may capture motion information to be included in a point cloud as an attribute associated with one or more points of the point cloud. For example, a vehicle equipped with a LIDAR system, a 3-D camera, or a 3-D scanner may include the vehicle’s direction and speed in a point cloud captured by the LIDAR system, the 3-D camera, or the 3-D scanner. For example, when points in a view of the vehicle are captured they may be included in a point cloud, where the point cloud includes the captured points and associated motion information corresponding to a state of the vehicle when the points were captured.

[0057] In some embodiments, the one or more patch images can include attribute and/or spatial information of the point cloud projected onto the patch image using one or more projections. For example, projections may include cylindrical or spherical projections, where the point cloud is projected onto a cylinder or sphere. Also, in some embodiments, multiple parallel projections of the point cloud may be used to generate patch images for the point cloud, where the multiple projections are known by or signaled to a decoder. In some implementations, one or more patch images can be packed in to one or more image frames of a video. The image frames can be encoded according to a video encoding standard, such as the high efficiency video coding (HEVC) standard or some other image or video coding standard or specification (e.g., VP9, VP10, or some other standard or specification).

[0058] In some embodiments, attribute and/or spatial information for a point cloud can be compressed by projecting the point cloud onto multiple projections and encoding the projections (e.g., in one or more layers of a patch image). For example, projections may include cylindrical or spherical projections, where the point cloud is projected onto a cylinder or sphere. Also, in some embodiments, multiple parallel projections of the point cloud may be encoded, where the multiple projections are known by or signaled to a decoder.

[0059] In some embodiments, points of a point cloud may be in a same or nearly same location when projected onto a patch plane. For example, the point cloud might have a depth such that some points are in the same location relative to the patch plane, but at different depths. An occupancy map having one or more layers can be generated to provide information regarding one or more of these points. For example, an occupancy map can indicate, for each image frame, the locations of one or more patch images packed into the image frame, and depth information of one or more sets of points corresponding to the patch images in the image frame. Further, the depth information can indicate, for each patch image, depths of the set of points corresponding to the patch image (e.g., with respect to a projection direction perpendicular to the patch plane of the patch image).

Example System Arrangement

[0060] FIG. 1 illustrates a system including a sensor that captures information for points of a point cloud and an encoder that compresses attribute information of the point cloud, where the compressed attribute information is sent to a decoder.

[0061] System 100 includes sensor 102 and encoder 104. Sensor 102 captures a point cloud 110 including points representing structure 106 in view 108 of sensor 102. For example, in some embodiments, structure 106 may be a mountain range, a building, a sign, an environment surrounding a street, or any other type of structure. In some embodiments, a captured point cloud, such as captured point cloud 110, may include spatial and attribute information for the points included in the point cloud. For instance, in the example shown in FIG. 1, point A of captured point cloud 110 includes X, Y, Z coordinates and attributes 1, 2, and 3. In some embodiments, attributes of a point may include attributes such as R, G, B color values, a velocity at the point, an acceleration at the point, a reflectance of the structure at the point, a time stamp indicating when the point was captured, a string-value indicating a modality when the point was captured, for example “walking,” or other attributes. The captured point cloud 110 may be provided to encoder 104, where encoder 104 generates a compressed version of the point cloud (e.g., compressed attribute information 112) that is transmitted via network 114 to decoder 116. In some embodiments, a compressed version of the point cloud, such as compressed attribute information 112, may be included in a common compressed point cloud that also includes compressed spatial information for the points of the point cloud or, in some embodiments, compressed spatial information and compressed attribute information may be communicated as separate sets of data.

[0062] In some embodiments, encoder 104 may be integrated with sensor 102. For example, encoder 104 may be implemented in hardware or software included in a sensor device, such as sensor 102. In other embodiments, encoder 104 may be implemented on a separate computing device that is proximate to sensor 102.

Example Intra-Frame Encoder

[0063] FIG. 2A illustrates components of an encoder for encoding intra point cloud frames. In some embodiments, the encoder described above in regard to FIG. 1 may operate in a similar manner as encoder 200 described in FIG. 2A.

[0064] The encoder 200 receives uncompressed point cloud 202 and generates compressed point cloud information 204. In some embodiments, an encoder, such as encoder 200, may receive the uncompressed point cloud 202 from a sensor, such as sensor 102 illustrated in FIG. 1, or, in some embodiments, may receive the uncompressed point cloud 202 from another source, such as a graphics generation component that generates the uncompressed point cloud in software, as an example.

[0065] In some embodiments, an encoder, such as encoder 200, includes decomposition into patches module 206, packing module 208, an image frame padding module 210, video compression module 212, and multiplexer 214. In addition, an encoder can include a patch information compression module, such as patch information compression module 216.

[0066] In some embodiments, the conversion process decomposes the point cloud into a set of patches (e.g., a patch is defined as a contiguous subset of the surface described by the point cloud), which may be overlapping or not, such that each patch may be described by a depth field with respect to a plane in 2D space. More details about the patch decomposition process are provided above with regard to FIGS. 3A-3C. Further, the encoder can produce one or more of geometry information, attribute information, and/or occupancy map information regarding the point cloud.

[0067] After or in conjunction with the patches being determined for the point cloud being compressed, a 2D sampling process is performed in planes associated with the patches. The 2D sampling process may be applied in order to approximate each patch with a uniformly sampled point cloud, which may be stored as a set of 2D patch images describing the occupancy map, geometry, and/or attributes of the point cloud at the patch location. The “packing” module 208 may store the 2D patch images associated with the patches in a single (or multiple) 2D images, referred to herein as “image frames.” In some embodiments, a packing module, such as packing module 208, may pack the 2D patch images such that the packed 2D patch images do not overlap (even though an outer bounding box for one patch image may overlap an outer bounding box for another patch image). Also, the packing module 208 may pack the 2D patch images in a way that minimizes non-used images pixels of the image frame. In some implementations, patch information can be used to convert the projected images into a point cloud by indicating sizes and shapes of the patches, the locations of the patches, and/or other information regarding the patches. This information can be encoded by a patch-information compression module, such as patch information compression module 216.

[0068] In some embodiments, 2D patch images associated with the occupancy map, geometry, and/or attributes of a point cloud can be generated at a given patch location. As noted before, a packing process, such as performed by packing module 208, may leave some empty spaces between 2D patch images packed in an image frame. Also, a padding module, such as image frame padding module 210, may fill in such areas in order to generate an image frame that may be suited for 2D video and image codecs.

[0069] In some embodiments, an occupancy map (e.g., information describing for each pixel or block of pixels whether the pixel or block of pixels are padded or not, and depth information for one or more points associated with that pixel or block of pixels) may be generated and compressed. The occupancy map may be sent to a decoder to enable the decoder to distinguish between padded and non-padded pixels of an image frame, and to determining the depth of one or more points associated with the padded pixels of the image frame.

[0070] In some embodiments one or more image frames may be encoded by a video encoder, such as video compression module 212. In some embodiments, a video encoder, such as video compression module 212, may operate in accordance with the High Efficiency Video Coding (HEVC) standard or other suitable video encoding standard or specification (e.g., VP9, VP10, or some other standard or specification). In some embodiments, encoded video images, encoded occupancy map information, and encoded patch information may be multiplexed by a multiplexer, such as multiplexer 214, and provided to a recipient as compressed point cloud information, such as compressed point cloud information 204.

[0071] In some embodiments, an occupancy map may be encoded and decoded by a video compression module, such as video compression module 212. This may be done at an encoder, such as encoder 200, such that the encoder has an accurate representation of what the occupancy map will look like when decoded by a decoder. Also, variations in image frames due to lossy compression and decompression may be accounted for when determining an occupancy map for an image frame. In some embodiments, various techniques may be used to further compress an occupancy map, such as described in FIGS. 7-11.

Example Intra-Frame Decoder

[0072] FIG. 2B illustrates components of a decoder for decoding intra point cloud frames. Decoder 230 receives compressed point cloud information 204, which may be the same compressed point cloud information 204 generated by encoder 200. Decoder 230 generates reconstructed point cloud 246 based on receiving the compressed point cloud information 204.

[0073] In some embodiments, a decoder, such as decoder 230, includes a de-multiplexer 232, a video decompression module 234, and an patch-information decompression module 238. Additionally a decoder, such as decoder 230 includes a point cloud generation module 240, which reconstructs a point cloud based on patch images included in one or more image frames included in the received compressed point cloud information, such as compressed point cloud information 204. In some embodiments, a decoder, such as decoder 230, further includes a smoothing filter, such as smoothing filter 244. In some embodiments, a smoothing filter may smooth incongruences at edges of patches, where data included in patch images for the patches has been used by the point cloud generation module to recreate a point cloud from the patch images for the patches. In some embodiments, a smoothing filter may be applied to the pixels located on the patch boundaries to alleviate the distortions that may be caused by the compression/decompression process.

Segmentation Process

[0074] FIG. 3A illustrates an example segmentation process for determining patches for a point cloud. The segmentation process as described in FIG. 3A may be performed by a decomposition into patches module, such as decomposition into patches module 206. A segmentation process may decompose a point cloud into a minimum number of patches (e.g., a contiguous subset of the surface described by the point cloud), while making sure that the respective patches may be represented by a depth field with respect to a patch plane. This may be done without a significant loss of shape information.

[0075] In some embodiments, a segmentation process may include: [0076] Letting point cloud PC be the input point cloud to be partitioned into patches and {P(0), P(1) … , P(N-1)} be the positions of points of point cloud PC. [0077] In some embodiments, a fixed set D={D(0), D(1), … , D (K-1)} of K 3D orientations is pre-defined. For instance, D may be chosen as follows D={(1.0, 0.0, 0.0), (0.0, 1.0, 0.0), (0.0, 0.0, 1.0), (-1.0, 0.0, 0.0), (0.0, -1.0, 0.0), (0.0, 0.0, -1.0)}. [0078] In some embodiments, the normal vector to the surface at every point P(i) is estimated. Any suitable algorithm may be used to determine the normal vector to the surface. For instance, a technique could include fetching the set H of the N nearest points of P(i), and fitting a plane .PI.(i) to H(i) by using principal component analysis techniques. The normal to P(i) may be estimated by taking the normal V(i) to .PI.(i). Note that N may be a user-defined parameter or may be found by applying an optimization procedure. N may also be fixed or adaptive. The normal values may then be oriented consistently by using a minimum-spanning tree approach. [0079] Normal-based Segmentation: An initial segmentation S0 of the points of point cloud PC may be obtained by associating respective points with the direction D(k) which maximizes the score (.A-inverted.(i)|D(k)), where (.|.) is the canonical dot product of R3. Pseudo code is provided below:

TABLE-US-00001 [0079] TABLE 1 Pseudo code for normal-based segmentation. for (i = O; i < pointCount; ++i) { clusterlndex = 0; bestScore = <.gradient.(i)|D(0)>; for(j = 1; j < K; ++j) { score= <.gradient.(i)|D(j)>; if (score> bestScore) { bestScore = score; clusterlndex = j; } } partition[i] = clusterIndex; }

[0080] Iterative segmentation refinement: Note that segmentation S0 associates respective points with the plane .PI.(i) that best preserves the geometry of its neighborhood (e.g., the neighborhood of the segment). In some circumstances, segmentation S0 may generate too many small connected components with irregular boundaries, which may result in poor compression performance. In order to avoid such issues, the following iterative segmentation refinement procedure may be applied: [0081] 1. An adjacency graph A may be built by associating a vertex V(i) to respective points P(i) of point cloud PC and by adding R edges {E(i,j(0)), … , E(i, j(R-1)} connecting vertex V(i) to its nearest neighbors {V(j(0)), V(j(1)), … , V(j(R-1))}. More precisely, {V(j(0)), V(j(1)), … , V(j(R-1))} may be the vertices associated with the points {P(j(0)), P(j(1)), … , P(j(R-1))}, which may be the nearest neighbors of P(i). Note that R may be a user-defined parameter or may be found by applying an optimization procedure. It may also be fixed or adaptive. [0082] 2. At each iteration, the points of point cloud PC may be traversed and every vertex may be associated with the direction D(k) that maximizes

[0082] ( .gradient. ( i ) | D ( k ) + .lamda. R .zeta. ( i ) ) , ##EQU00001## where .zeta.(i) is the number of the R-nearest neighbors of V(i) belonging to the same cluster and .lamda. is a parameter controlling the regularity of the produced patches. Note that the parameters .lamda. and R may be defined by the user or may be determined by applying an optimization procedure. They may also be fixed or adaptive. In some embodiments, a “user” as referred to herein may be an engineer who configured a point cloud compression technique as described herein to one or more applications. [0083] 3. An example of pseudo code is provided below:

TABLE-US-00002 [0083] TABLE 2 Pseudo code for iterative segmentation refinement. for(I = 0; I< iterationCount; ++I) { for(i = O; i < pointCount; ++i) { clusterlndex = partition[i]; bestScore = 0.0; for(k = 0; k < K; ++k) { score= .gradient.(i)|D(k) ; for(j .di-elect cons. {j(0),j(1), … ,j(R-1)}) { if (k == partition[j]) { score += .lamda. / R’ ; } } if (score> bestScore) { bestScore = score; clusterlndex = k; } } partition[i] = clusterIndex;

[0084] In some embodiments, the pseudo code shown above may further include an early termination step. For example, if a score that is a particular value is reached, or if a difference between a score that is reached and a best score only changes by a certain amount or less, the search could be terminated early. Also, the search could be terminated if after a certain number of iterations (l=m), the clusterindex does not change. [0085] Patch segmentation: In some embodiments, the patch segmentation procedure further segments the clusters detected in the previous steps into patches, which may be represented with a depth field with respect to a projection plane. The approach proceeds as follows, according to some embodiments: [0086] 1. First, a cluster-based adjacency graph with a number of neighbors R’ is built, while considering as neighbors only the points that belong to the same cluster. Note that R’ may be different from the number of neighbors R used in the previous steps. [0087] 2. Next, the different connected components of the cluster-based adjacency graph are extracted. Only connected components with a number of points higher than a parameter a are considered. Let CC={CC(0), CC(1), … , CC(M-1)} be the set of the extracted connected components. [0088] 3. Respective connected component CC(m) inherits the orientation D(m) of the cluster it belongs to. The points of CC(m) are then projected on a projection plane having as normal the orientation D(m), while updating a depth map, which records for every pixel the depth of the nearest point to the projection plane. [0089] 4. An approximated version of CC(m), denoted C’(m), is then built by associating respective updated pixels of the depth map with a 3D point having the same depth. Let PC be the point cloud obtained by the union of reconstructed connected components {CC’(0), CC’(1), … , CC’(M-1)}. [0090] 5. Note that the projection reconstruction process may be lossy and some points may be missing. In order, to detect such points, every point P(i) of point cloud PC may be checked to make sure it is within a distance lower than a parameter .delta. from a point of PC. If this is not the case, then P(i) may be marked as a missed point and added to a set of missed points denoted MP. [0091] 6. The steps 2-5 are then applied to the missed points MP. The process is repeated until MP is empty or CC is empty. Note that the parameters .delta. and a may be defined by the user or may be determined by applying an optimization procedure. They may also be fixed or adaptive. [0092] 7. A filtering procedure may be applied to the detected patches in order to make them better suited for compression. Example filter procedures may include: [0093] a. A smoothing filter based on the geometry/texture/attributes of the points of the patches (e.g., median filtering), which takes into account both spatial and temporal aspects. [0094] b. Discarding small and isolated patches. [0095] c. User-guided filtering. [0096] d. Other suitable smoothing filter techniques.

Layers

[0097] The image generation process described above includes projecting the points belonging to each patch onto its associated projection plane to generate a patch image. This process could be generalized to handle the situation where multiple points are projected onto the same pixel as follows: [0098] Let H(u, v) be the set of points of the current patch that get projected to the same pixel (u, v). Note that H(u, v) may be empty, may have one point or multiple points. [0099] If H(u, v) is empty then the pixel is marked as unoccupied. [0100] If the H(u, v) has a single element, then the pixel is filled with the associated geometry/texture/attribute value. [0101] If H(u, v), has multiple elements, then different strategies are possible: [0102] Keep only the nearest point P0(u, v) for the pixel (u, v) [0103] Take the average or a linear combination of a group of points that are within a distance d from P0(u, v), where d is a user-defined parameter needed only on the encoder side. [0104] Store two images: one for P0(u, v) and one to store the farthest point Pl(u, v) of H(u, v) that is within a distance d from P0(u, v) [0105] Store N patch images containing a subset of H(u, v)

[0106] The generated patch images for point clouds with points at the same patch location, but different depths may be referred to as layers herein. In some embodiments, scaling/up-sampling/down-sampling could be applied to the produced patch images/layers in order to control the number of points in the reconstructed point cloud.

[0107] Guided up-sampling strategies may be performed on the layers that were down-sampled given the full resolution image from another “primary” layer that was not down-sampled.

[0108] Down-sampling could leverage the closed loop techniques as described below in regard to closed-loop color conversion, while exploiting a guided up-sampling strategy. For example, a generated layer may be encoded independently, which allows for parallel decoding and error resilience. Also encoding strategies, such as those specified by the scalable-HEVC standard, may be leveraged in order to support advanced functionalities such as spatial, SNR (signal to noise ratio), and color gamut scalability.

[0109] In some embodiments, a delta prediction between layers could be adaptively applied based on a rate-distortion optimization. This choice may be explicitly signaled in the bit stream.

[0110] In some embodiments, the generated layers may be encoded with different precisions. The precision of each layer may be adaptively controlled by using a shift+scale or a more general linear or non-linear transformation.

[0111] In some embodiments, an encoder may make decisions on a scaling strategy and parameters, which are explicitly encoded in the bit stream. The decoder may read the information from the bit stream and apply the right scaling process with the parameters signaled by the encoder.

[0112] In some embodiments, a video encoding motion estimation process may be guided by providing a motion vector map to the video encoder indicating for each block of the image frame, a 2D search center or motion vector candidates for the refinement search. Such information, may be trivial to compute since the mapping between the 3D frames and the 2D image frames is available to the point cloud encoder and a coarse mapping between the 2D image frames could be computed by using a nearest neighbor search in 3D.

[0113] The video motion estimation/mode decision/intra-prediction could be accelerated/improved by providing a search center map, which may provide guidance on where to search and which modes to choose from for each N.times.N pixel block.

[0114] Hidden/non-displayed pictures could be used in codecs such as AV1 and HEVC. In particular, synthesized patches could be created and encoded (but not displayed) in order to improve prediction efficiency. This could be achieved by re-using a subset of the padded pixels to store synthesized patches.

[0115] The patch re-sampling (e.g., packing and patch segmentation) process described above exploits solely the geometry information. A more comprehensive approach may take into account the distortions in terms of geometry, texture, and other attributes and may improve the quality of the re-sampled point clouds.

[0116] Instead of first deriving the geometry image and optimizing the texture image given said geometry, a joint optimization of geometry and texture could be performed. For example, the geometry patches could be selected in a manner that results in minimum distortion for both geometry and texture. This could be done by immediately associating each possible geometry patch with its corresponding texture patch and computing their corresponding distortion information. Rate-distortion optimization could also be considered if the target compression ratio is known.

[0117] In some embodiments, a point cloud resampling process described above may additionally consider texture and attributes information, instead of relying only on geometry.

[0118] Also, a projection-based transformation that maps 3D points to 2D pixels could be generalized to support arbitrary 3D to 2D mapping as follows: [0119] Store the 3D to 2D transform parameters or the pixel coordinates associated with each point [0120] Store X, Y, Z coordinates in the geometry images instead of or in addition to the depth information

Packing

[0121] In some embodiments, depth maps associated with patches, also referred to herein as “depth patch images,” such as those described above, may be packed into a 2D image frame. For example, a packing module, such as packing module 208, may pack depth patch images generated by a spatial image generation module. The depth maps, or depth patch images, may be packed such that (A) no non-overlapping block of T.times.T pixels contains depth information from two different patches and such that (B) a size of the generated image frame is minimized.

[0122] In some embodiments, packing includes the following steps: [0123] a. The patches are sorted by height and then by width. The patches are then inserted in image frame (I) one after the other in that order. At each step, the pixels of image frame (I) are traversed in raster order, while checking if the current patch could be inserted under the two conditions (A) and (B) described above. If it is not possible then the height of (I) is doubled. [0124] b. This process is iterated until all the patches are inserted.

[0125] In some embodiments, the packing process described above may be applied to pack a subset of the patches inside multiples tiles of an image frame or multiple image frames. This may allow patches with similar/close orientations based on visibility according to the rendering camera position to be stored in the same image frame/tile, to enable view-dependent streaming and/or decoding. This may also allow parallel encoding/decoding.

[0126] In some embodiments, the packing process can be considered a bin-packing problem and a first decreasing strategy as described above may be applied to solve the bin-packing problem. In other embodiments, other methods such as the modified first fit decreasing (MFFD) strategy may be applied in the packing process.

[0127] In some embodiments, if temporal prediction is used, such as described for inter compression encoder 250, such an optimization may be performed with temporal prediction/encoding in addition to spatial prediction/encoding. Such consideration may be made for the entire video sequence or per group of pictures (GOP). In the latter case additional constraints may be specified. For example, a constraint may be that the resolution of the image frames should not exceed a threshold amount. In some embodiments, additional temporal constraints may be imposed, even if temporal prediction is not used, for example such as that a patch corresponding to a particular object view is not moved more than x number of pixels from previous instantiations.

[0128] FIG. 3B illustrates an example image frame including packed patch images and padded portions. Image frame 300 includes patch images 302 packed into image frame 300 and also includes padding 304 in space of image frame 300 not occupied by patch images. In some embodiments, padding, such as padding 304, may be determined so as to minimize incongruences between a patch image and the padding. For example, in some embodiments, padding may construct new pixel blocks that are replicas of, or are to some degree similar to, pixel blocks that are on the edges of patch images. Because an image and/or video encoder may encode based on differences between adjacent pixels, such an approach may reduce the number of bytes required to encode an image frame including patch images and padding.

[0129] In some embodiments, the patch information may be stored in the same order as the order used during the packing, which makes it possible to handle overlapping 2D bounding boxes of patches. Thus, a decoder receiving the patch information can extract patch images from the image frame in the same order in which the patch images were packed into the image frame. Also, because the order is known by the decoder, the decoder can resolve patch image bounding boxes that overlap.

[0130] FIG. 3C illustrates an example image frame 312 with overlapping patches, according to some embodiments. FIG. 3C shows an example with two patches (patch image 1 and patch image 2) having overlapping 2D bounding boxes 314 and 316 that overlap at area 318. In order to determine to which patch the T.times.T blocks in the area 318 belong, the order of the patches may be considered. For example, the T.times.T block 314 may belong to the last decoded patch. This may be because in the case of an overlapping patch, a later placed patch is placed such that it overlaps with a previously placed patch. By knowing the placement order it can be resolved that areas of overlapping bounding boxes go with the latest placed patch. In some embodiments, the patch information is predicted and encoded (e.g., with an entropy/arithmetic encoder). Also, in some embodiments, U0, V0, DUO and DV0 are encoded as multiples of T, where T is the block size used during the padding phase.

[0131] FIG. 3C also illustrates blocks of an image frame 312, where the blocks may be further divided into sub-blocks. For example block A1, B1, C1, A2, etc. may be divided into multiple sub-blocks, and, in some embodiments, the sub-blocks may be further divided into smaller blocks. In some embodiments, a video compression module of an encoder, such as video compression module 212 or video compression module 264, may determine whether a block includes active pixels, non-active pixels, or a mix of active and non-active pixels. The video compression module may budget fewer resources to compress blocks including non-active pixels than an amount of resources that are budgeted for encoding blocks including active pixels. In some embodiments, active pixels may be pixels that include data for a patch image and non-active pixels may be pixels that include padding. In some embodiments, a video compression module may sub-divide blocks including both active and non-active pixels, and budget resources based on whether sub-blocks of the blocks include active or non-active pixels. For example, blocks A1, B1, C1, A2 may include non-active pixels. As another example block E3 may include active pixels, and block B6, as an example, may include a mix of active and non-active pixels.

[0132] In some embodiments, a patch image may be determined based on projections, such as projecting a point cloud onto a cube, cylinder, sphere, etc. In some embodiments, a patch image may include a projection that occupies a full image frame without padding. For example, in a cubic projection each of the six cubic faces may be a patch image that occupies a full image frame.

[0133] For example, FIG. 3D illustrates a point cloud being projected onto multiple projections.

[0134] In some embodiments, a representation of a point cloud is encoded using multiple projections. For example, instead of determining patches for a segment of the point cloud projected on a plane perpendicular to a normal to the segment, the point cloud may be projected onto multiple arbitrary planes or surfaces. For example, a point cloud may be projected onto the sides of a cube, cylinder, sphere, etc. Also multiple projections intersecting a point cloud may be used. In some embodiments, the projections may be encoded using conventional video compression methods, such as via a video compression module 212 or video compression module 264. In particular, the point cloud representation may be first projected onto a shape, such as a cube, and the different projections/faces projected onto that shape (i.e., front (320), back (322), top (324), bottom (326), left (328), right (330)) may all be packed onto a single image frame or multiple image frames. This information, as well as depth information may be encoded separately or with coding tools such as the ones provided in the 3D extension of the HEVC (3D-HEVC) standard. The information may provide a representation of the point cloud since the projection images can provide the (x,y) geometry coordinates of all projected points of the point cloud. Additionally, depth information that provides the z coordinates may be encoded. In some embodiments, the depth information may be determined by comparing different ones of the projections, slicing through the point cloud at different depths. When projecting a point cloud onto a cube, the projections might not cover all point cloud points, e.g., due to occlusions. Therefore, additional information may be encoded to provide for these missing points and updates may be provided for the missing points.

[0135] In some embodiments, adjustments to a cubic projection can be performed that further improve upon such projections. For example, adjustments may be applied at the encoder only (non-normative) or applied to both the encoder and the decoder (normative).

[0136] More specifically, in some embodiments alternative projections may be used. For example, instead of using a cubic projection, a cylindrical or spherical type of a projection method may be used. Such methods may reduce, if not eliminate, redundancies that may exist in the cubic projection and reduce the number or the effect of “seams” that may exist in cubic projections. Such seams may create artifacts at object boundaries, for example. Eliminating or reducing the number or effect of such seams may result in improved compression/subjective quality as compared to cubic projection methods. For a spherical projection case, a variety of sub-projections may be used, such as the equirectangular, equiangular, and authagraph projection among others. These projections may permit the projection of a sphere onto a 2D plane. In some embodiments, the effects of seams may be de-emphasized by overlapping projections, where multiple projections are made of a point cloud, and the projections overlap with one another at the edges, such that there is overlapping information at the seams. A blending effect could be employed at the overlapping seams to reduce the effects of the seams, thus making them less visible.

[0137] In addition to, or instead of, considering a different projection method (such as cylindrical or spherical projections), in some embodiments multiple parallel projections may be used. The multiple parallel projections may provide additional information and may reduce a number of occluded points. The projections may be known at the decoder or signaled to the decoder. Such projections may be defined on planes or surfaces that are at different distances from a point cloud object. Also, in some embodiments the projections may be of different shapes, and may also overlap or cross through the point cloud object itself. These projections may permit capturing some characteristics of a point cloud object that may have been occluded through a single projection method or a patch segmentation method as described above.

[0138] For example, FIG. 3E illustrates a point cloud being projected onto multiple parallel projections, according to some embodiments. Point cloud 350 which includes points representing a coffee mug is projected onto parallel horizontal projections 352 that include planes orthogonal to the Z axis. Point cloud 350 is also projected onto vertical projections 354 that include planes orthogonal to the X axis, and is projected onto vertical projections 356 that include planes orthogonal to the Y axis. In some embodiments, instead of planes, multiple projections may include projections having other shapes, such as multiple cylinders or spheres.

Generating Images Having Depth

[0139] In some embodiments, only a subset of the pixels of an image frame will be occupied and may correspond to a subset of 3D points of a point cloud. Information regarding the points (e.g., geometry, texture, and other attributes) can be encoded by generating maps corresponding to the patch images, and storing, for each occupied pixel in the map, the depth/texture/attribute value of its associated point(s) of the patch images.

[0140] In some embodiments, spatial information may be stored with various variations, for example spatial information may: [0141] a. Store depth as a monochrome image. [0142] b. Store depth as Y and keep U and V empty (where YUV is a color space, also RGB color space may be used). [0143] c. Store depth information for different patches in different color planes Y, U and V, in order to avoid inter-patch contamination during compression and/or improve compression efficiency (e.g., have correlated patches in the same color plane). Also, hardware codec capabilities may be utilized, which may spend the same encoding/decoding time independently of the content of the frame. [0144] d. Store depth patch images on multiple images or tiles that could be encoded and decoded in parallel. One advantage is to store depth patch images with similar/close orientations or based on visibility according to the rendering camera position in the same image/tile, to enable view-dependent streaming and/or decoding. [0145] e. Store depth as Y and store a redundant version of depth in U and V. [0146] f. Store X, Y, Z coordinates in Y, U, and V. [0147] g. Different bit depth (e.g., 8, 10 or 12-bit) and sampling (e.g., 420, 422, 444 … ) may be used. Note that different bit depth may be used for the different color planes. [0148] h. Generate an occupancy map having one or more layers. The occupancy map can indicate, for each occupied pixel of an image frame, the depth/texture/attribute value of its associated point(s). For example, an occupancy map can indicate, for each image frame, the locations of one or more patch images packed into the image frame, and depth information of one or more sets of points corresponding to the patch images in the image frame. Further, the depth information can indicate, for each patch image, depths of the set of points corresponding to the patch image (e.g., with respect to a projection direction perpendicular to the patch plane of the patch image). Example techniques for generating occupancy maps are shown and described with respect to FIGS. 4-9 and 11. [0149] i. Store one or more additional images (e.g., in conjunction with one or more patch images and/or occupancy maps), each containing attribute information regarding points of the point cloud (e.g., color information or other attribute information regarding occluded points). Example techniques for generating additional images containing attribute information are shown and described with respect to FIG. 6.

Padding

[0150] In some embodiments, padding may be performed to fill the non-occupied pixels with values such that the resulting image is suited for video/image compression. For example, image frame padding module 210 or image padding module 262 may perform padding as described below.

[0151] In some embodiments, padding is applied on pixels blocks, while favoring the intra-prediction modes used by existing video codecs. More precisely, for each block of size B.times.B to be padded, the intra prediction modes available at the video encoder side are assessed and the one that produces the lowest prediction errors on the occupied pixels is retained. This may take advantage of the fact that video/image codecs commonly operate on pixel blocks with pre-defined sizes (e.g., 64.times.64, 32.times.32, 16.times.16 … ). In some embodiments, other padding techniques may include linear extrapolation, in-painting techniques, or other suitable techniques.

Video Compression

[0152] In some embodiments, a video compression module, such as video compression module 212 or video compression module 264, may perform video compression as described below.

[0153] In some embodiments, a video encoder may leverage an occupancy map, which describes for each pixel of an image whether it stores information belonging to the point cloud or padded pixels (among other information). In some embodiments, such information may permit enabling various features adaptively, such as de-blocking, adaptive loop filtering (ALF), or shape adaptive offset (SAO) filtering. Also, such information may allow a rate control module to adapt and assign different, e.g., lower, quantization parameters (QPs), and in an essence a different amount of bits, to the blocks containing the occupancy map edges. Coding parameters, such as lagrangian multipliers, quantization thresholding, quantization matrices, etc. may also be adjusted according to the characteristics of the point cloud projected blocks. In some embodiments, such information may also enable rate distortion optimization (RDO) and rate control/allocation to leverage the occupancy map to consider distortions based on non-padded pixels. In a more general form, weighting of distortion may be based on the “importance” of each pixel to the point cloud geometry. Importance may be based on a variety of aspects, e.g., on proximity to other point cloud samples, directionality/orientation/position of the samples, etc. Facing forward samples, for example, may receive a higher weighting in the distortion computation than backward facing samples. Distortion may be computed using metrics such as Mean Square or Absolute Error, but different distortion metrics may also be considered, such as SSIM, VQM, VDP, Hausdorff distance, and others.

Point Cloud Compression

[0154] As described herein, a point cloud can be represented by one or more videos each having one or more image frames, where each image frame is packed with one or more patch images, and where each occupied pixel of an image frame corresponds to one or more respective 3D points in the point cloud. Further, information regarding the points (e.g., geometry and other attributes) can be encoded by generating maps corresponding to the patch images, and storing, for each occupied pixel in the map, the depth, and other attribute value of its associated point(s) of the patch images. Each of these maps can be stored as one or more image frames in a video.

[0155] In some implementations, information regarding the points (e.g., geometry) can encoded in one or more geometry images and/or occupancy maps. In some implementations, the geometry images and/or occupancy maps can be stored as one or more image frames of one or more videos. As an example, the geometry images can be stored as one or more image frames of a first video, and an occupancy map can be stored as one or more images frames of a second video.

[0156] In some implementations, an occupancy map can indicate the presence and location of a point with respect to a projection plane (e.g., the presence of a point in a direction perpendicular to the projection plane). In some implementations, an occupancy map can indicate the presence and locations of multiple points with respect to a projection plane, including points that are occluded by other points with respect to the projection plane. For example, the occupancy map can indicate not only the presence and location of the point nearest to the projection plane in a direction perpendicular to the projection plane (e.g., the depth of the point), but also the presence and locations of one or more additional points farther from the projection plane in the direction perpendicular to the projection plane (e.g., the depth of the points occluded by the nearest point).

[0157] Further, an occupancy map can be compressed. This can be beneficial, for example, in reducing costs and time associated with storing, processing, and/or transmitting data regarding the point cloud. In some implementations, an occupancy may can down-sampled and/or encoded in a lossy manner (e.g., such that at least some of the information of the occupancy map is discarded). In some implementations, a down-sampled and/or encoded occupancy map can be subsequently reconstructed, such that least some of the discarded information is recovered).

[0158] FIG. 4 illustrates an example process of generating an occupancy map 400 and a geometry image 410 representing one or more points in a point cloud 402. In this example, a point cloud 402 includes a number of 3D points 404 (represented by shaded boxes in a grid). For ease of illustration, FIG. 4 depicts the points 404 on a single plane of the point cloud 402 (e.g., a single x-y plane). However, in practice, the point cloud 402 can include multiple points 404 on multiple different planes (e.g., multiple x-y planes stacked along the z-direction). In some implementations, the point cloud 402 can be included in and/or represent three-dimensional visual volumetric content.

[0159] As described herein, the points 404 of the point cloud 402 can be projected onto 2D planes in one or more groups, and stored as one or more 2D images (e.g., patch images). Further, multiple points 404 may end up being projected onto the same position of the planes. In the example shown in FIG. 4, the points 404 are projected in a projection direction (e.g., in the negative y direction) onto a projection plane 408. Due to the arrangement of the points 404, at least some points are occluded by other points with respect to the projection plane 408 (the bottommost point in each column of the grid occludes one or more other points above it in the column).

[0160] An occupancy map 400 and geometry image 410 can be generated to provide information regarding one or more of the points 404 in the point cloud 402. For instance, a geometry image 410 can include one or more layers, each indicating certain information regarding one or more of the points 404. Further, an occupancy map 400 can indicate additional information regarding one or more of the points 404.

[0161] As an example, the geometry image 410 can include a first layer 410a indicating the depth of the point 404 nearest to the projection plane 408 (represented by shaded boxes marked “D0”) minus the minimum depth across the columns (e.g., in this example, 1). For instance, proceeding from the left column to the right column, the values in the first layer 410a are 1, 2, 3, 3, null (as there are no points in the column), 1, 0, and 1, respectively.

[0162] As another example, the geometry image 410 can include a second layer 410b indicating the depth of additional points 404 farther from the projection plane 408. In some implementations, the second layer 400b can indicate the depth of the farthest point 404 from the projection plane 408 in a column, within a particular surface thickness t.sub.surface from the nearest point 404 in the column (represented by shaded boxed marked “D1”), and minus the minimum depth across the columns. For instance, in the example shown in FIG. 4, the surface thickness t.sub.surface is 4, and is indicated in each column by a thick horizontal line. Proceeding from the left column to the right column, values of the second layer 410b are 4, 5, 7, null (as there are no additional points in that column), null (as there are no points in that column), null (as there are no points within 4 of the nearest point in that column), 2, and 4, respectively. Although a surface thickness t.sub.surface of 4 is shown in FIG. 4, this is merely an illustrative example. In practice, the surface thickness t.sub.surface can vary, depending on the implementation. In some implementations, the surface thickness t.sub.surface can be selected empirically by a user (e.g., based on the requirements for a particular application).

[0163] Further, the occupancy map 400 indicates whether at least one point has been projected onto a particular location of the projection plane 408. For instance, proceeding from the left column to the right column, this can indicated as 1 (indicating that at least one point has been projected onto the projection plane 408 with respect to that column), 1, 1, 1, 0 (indicating that no points have been projected onto the projection plane 408 with respect to that column), 1, 1, and 1, respectively.

[0164] The points 404 that are not represented by the occupancy map 400 (e.g., the points 404 (i) between the nearest point and the farthest point within a particular surface thickness from the nearest point, represented by shaded boxes marked “o,” and/or (ii) the points beyond the surface thickness from the nearest point, represented by shaped boxed marked “x”), and/or (iii) the points the encoder decides not to project can be encoded in one of more other images and/or image layers. For example, the remaining points 404 can be encoded by explicitly signal the geometry values and stored as one or more additional patch images (e.g., in one or more image frames of a video).

[0165] However, in some implementations, it may be less desirable to encode information regarding the points 404 according to explicitly signaling, due to the computational resources and/or time needed to generate, store, and/or transmit information encoded in this manner. As an alternative, information regarding at least some of the points 404 can be encoded according to alternative techniques, rather than according to explicitly signaling.

……
……
……