WO2022166957A1 - 点云数据的预处理方法及点云几何编解码方法、装置 - Google Patents

点云数据的预处理方法及点云几何编解码方法、装置 Download PDF

Info

Publication number
WO2022166957A1
WO2022166957A1 PCT/CN2022/075379 CN2022075379W WO2022166957A1 WO 2022166957 A1 WO2022166957 A1 WO 2022166957A1 CN 2022075379 W CN2022075379 W CN 2022075379W WO 2022166957 A1 WO2022166957 A1 WO 2022166957A1
Authority
WO
WIPO (PCT)
Prior art keywords
prediction
point
point cloud
geometric
regularized
Prior art date
Application number
PCT/CN2022/075379
Other languages
English (en)
French (fr)
Other versions
WO2022166957A9 (zh
Inventor
杨付正
张伟
Original Assignee
荣耀终端有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 荣耀终端有限公司 filed Critical 荣耀终端有限公司
Priority to EP22749236.0A priority Critical patent/EP4246442A4/en
Priority to US18/267,977 priority patent/US20240062429A1/en
Publication of WO2022166957A1 publication Critical patent/WO2022166957A1/zh
Publication of WO2022166957A9 publication Critical patent/WO2022166957A9/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/001—Model-based coding, e.g. wire frame
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/40—Tree coding, e.g. quadtree, octree
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/005—Tree description, e.g. octree, quadtree
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/06—Topological mapping of higher dimensional structures onto lower dimensional surfaces
    • G06T3/073—Transforming surfaces of revolution to planar images, e.g. cylindrical surfaces to planar images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/80—Geometric correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/004—Predictors, e.g. intraframe, interframe coding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/597—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding specially adapted for multi-view video sequence encoding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/85—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/56—Particle system, point based geometry or rendering
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/70—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by syntax aspects related to video coding, e.g. related to compression standards

Definitions

  • the invention belongs to the technical field of point cloud data processing, and in particular relates to a point cloud data preprocessing method and a point cloud geometry encoding and decoding method and device.
  • the geometric information of the point cloud and the attribute information corresponding to each point are encoded separately.
  • the geometric encoding and decoding of G-PCC can be divided into geometric encoding and decoding based on octree and geometric encoding and decoding based on prediction tree.
  • the geometric coding based on prediction tree first sorts the input point cloud, and builds the prediction tree structure by using two different ways at the coding end.
  • each node in the prediction tree is traversed, the geometric position information of the node is predicted by selecting different prediction modes, and the prediction residual is obtained, and the geometric prediction residual is quantized by using quantization parameters. Finally, through continuous iteration, the prediction residual, prediction tree structure and quantization parameters of the prediction tree node position information are encoded to generate a binary code stream.
  • Predictive tree coding based on lidar calibration information is a commonly used geometric coding method.
  • the collection points belonging to the same laser scanner should be regularly distributed in the cylindrical coordinate system.
  • the actual data presents a non-uniform distribution, resulting in poor correlation between data, and low prediction accuracy and coding efficiency.
  • the point cloud encoding and decoding technology based on the prediction tree only uses some parameters of the lidar device to establish a tree structure, which does not fully reflect the spatial correlation of the point cloud, which is not conducive to the prediction and entropy encoding of the point cloud. affect the coding efficiency.
  • the existing G-PCC method only determines the relationship between each point and the laser scanner through vertical correction, resulting in the need to introduce other variables to assist the encoding of horizontal information during encoding, thereby increasing the amount of information to be encoded. Reduced geometry encoding efficiency.
  • the present invention provides a point cloud geometry prediction encoding and decoding method and device based on a regularized structure.
  • the technical problem to be solved by the present invention is realized by the following technical solutions:
  • a point cloud data preprocessing method comprising:
  • Regularization preprocessing is performed on the two-dimensional structure based on the geometric distortion measure to obtain a regularized structure.
  • regularization preprocessing is performed on the two-dimensional structure based on the geometric distortion measure to obtain a regularized structure, including:
  • the two-dimensional structure is adjusted according to the point-to-surface geometric distortion measure to obtain a regularized structure.
  • the two-dimensional structure is adjusted according to a point-to-surface geometric distortion measure to obtain a regularized structure, including:
  • regularized preprocessing is performed on the two-dimensional structure based on the geometric distortion measure to obtain a regularized structure, further comprising:
  • the two-dimensional structure is adjusted according to a point-to-line geometric distortion measure to obtain a regularized structure.
  • Another embodiment of the present invention also provides a point cloud geometry encoding method, characterized in that it includes:
  • the original point cloud data is subjected to regularization preprocessing using the preprocessing method described in the above embodiment to obtain a regularized structure
  • the information to be coded is sequentially coded to obtain a geometric information code stream.
  • the prediction mode of each point in the regularized structure is determined, and the selected prediction mode is used to perform geometric prediction on each point to obtain the information to be encoded, including:
  • the geometric prediction residual is used as part of the information to be encoded.
  • geometric prediction is performed on each point in the prediction tree structure according to the selected prediction mode, and the geometric prediction residual of each point is obtained, including:
  • the predicted value j' of the azimuth angle of the current point is calculated according to the following formula:
  • j prev represents the predicted azimuth of the current point
  • n represents the number of points that need to be skipped between the parent node and the current point according to the scanning speed, and its prediction residual is for n' represents the number of points that need to be skipped for the coded nodes adjacent to the current point;
  • Cartesian coordinates (x, y, z) of the current point and the predicted Cartesian coordinates Perform differential prediction to obtain the prediction residuals (r x , r y , r z ) in the Cartesian coordinate system.
  • Another embodiment of the present invention also provides a point cloud geometry encoding device, comprising:
  • a first data acquisition module used for acquiring original point cloud data
  • a regularization module which performs regularized preprocessing on the original point cloud data to obtain a regularized structure
  • the first prediction module is used to determine the prediction mode of each point in the regularized structure, and utilize the selected prediction mode to perform geometric prediction on each point to obtain the information to be encoded;
  • the encoding module is used for sequentially encoding the information to be encoded to obtain a geometric information code stream.
  • Another embodiment of the present invention also provides a point cloud geometry decoding method, including:
  • the decoded data includes the prediction mode of the current node
  • the prediction residuals include prediction residuals in a cylindrical coordinate system and prediction residuals in a Cartesian coordinate system;
  • Point cloud reconstruction is performed according to the prediction residuals in the Cartesian coordinate system and the predicted Cartesian coordinates to obtain reconstructed point cloud data.
  • Yet another embodiment of the present invention also provides a point cloud geometry decoding device, comprising:
  • the second data acquisition module is used to acquire the geometric information code stream, and decode it to obtain decoded data; wherein, the decoded data includes the prediction mode of the current node;
  • the second prediction module is configured to perform geometric prediction on the current node according to the prediction mode to obtain prediction residuals; wherein the prediction residuals include prediction residuals in the cylindrical coordinate system and prediction residuals in the Cartesian coordinate system Difference;
  • the prediction tree reconstruction module is used for reconstructing the prediction tree structure according to the prediction residuals under the cylindrical coordinate system, and performing coordinate conversion on the points in the prediction tree structure to obtain the predicted Cartesian coordinates of the current point;
  • the point cloud reconstruction module reconstructs the geometric point cloud according to the geometric prediction value and the prediction residual of the current node, and obtains the reconstructed point cloud data.
  • the point cloud data processing method provided by the present invention performs regularized preprocessing on the original input point cloud, so that the point cloud presents a regular distribution in both the horizontal and vertical directions, which better reflects the spatial correlation of the point cloud. , so as to facilitate further processing of point cloud data;
  • the geometric distortion measure is used to adjust the two-dimensional structure, which ensures the quality of the point cloud model
  • the point cloud geometric encoding method provided by the present invention performs regular processing on the azimuth direction of the point cloud when preprocessing the point cloud data, so that when encoding, it is not necessary to use an additional code stream to encode the azimuth direction. auxiliary information, saving the code stream and improving the coding efficiency;
  • the present invention performs geometric coding on the regularized point cloud, by effectively utilizing the regularized structure, the horizontal and vertical directions are combined with each other to perform predictive coding, thereby improving the geometric coding efficiency.
  • FIG. 1 is a schematic diagram of a point cloud data preprocessing method provided by an embodiment of the present invention
  • FIG. 2 is a schematic structural diagram of a lidar provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of a distribution structure of original collected data provided by an embodiment of the present invention.
  • FIG. 4 is an expanded view of a cylindrical coordinate system provided by an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of point-to-surface interpolation processing provided by an embodiment of the present invention
  • FIG. 6 is a before and after comparison diagram of regularization processing provided by an embodiment of the present invention.
  • FIG. 7 is a schematic diagram of a point cloud geometry encoding method provided by an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of a point cloud geometry encoding device provided by an embodiment of the present invention.
  • FIG. 9 is a schematic diagram of a point cloud geometry decoding method provided by an embodiment of the present invention.
  • FIG. 10 is a schematic structural diagram of a point cloud geometry decoding apparatus provided by an embodiment of the present invention.
  • FIG. 1 is a schematic diagram of a point cloud data preprocessing method provided by an embodiment of the present invention, including:
  • S1 Perform coordinate transformation on the original point cloud data to obtain the representation of the original point cloud in the cylindrical coordinate system.
  • the original point cloud data usually consists of a set of three-dimensional space points, and each space point records its own geometric position information, as well as additional attribute information such as color, reflectance, and normal.
  • the geometric position information of the point cloud is generally represented based on the Cartesian coordinate system, that is, it is represented by the x, y, and z coordinates of the point.
  • Raw point cloud data can be obtained through 3D scanning equipment such as lidar, etc., or through public datasets provided by various platforms.
  • the acquired geometric position information of the original point cloud data is represented as (x, y, z) based on the Cartesian coordinate system. It should be noted that the representation method of the geometric position information of the original point cloud data is not limited to Cartesian coordinates.
  • row quantization processing and reordering processing may also be performed on it, so as to facilitate subsequent predictive coding.
  • FIG. 2 is a schematic structural diagram of a lidar provided by an embodiment of the present invention; a lidar is composed of multiple laser scanners.
  • line number refers to the laser scanners included in it.
  • the number of these laser scanners is distributed along both sides of the central axis of the lidar, and has different pitch angles ⁇ i , so that the spatial information of different objects in the vertical direction in space can be obtained.
  • ⁇ i the pitch angles
  • Each can be regarded as a relatively independent acquisition system. Through the rotation of the base during acquisition, all laser scanners acquire the position information of objects in space according to a certain sampling rate.
  • Cartesian coordinates (x, y, z) of the original point cloud data can be converted into cylindrical coordinates (r, ⁇ , i) according to the existing conversion formula, and the original point cloud on the cylindrical surface can be obtained. Representation in the coordinate system.
  • FIG. 3 is a schematic diagram of a distribution structure of original collected data in a cylindrical coordinate system provided by an embodiment of the present invention.
  • the point cloud data acquired by LiDAR presents a uniform distribution along the azimuth and elevation directions in a cylindrical coordinate system.
  • the point cloud presents the characteristics of non-uniform distribution.
  • the structure and acquisition parameters of the lidar are first used to determine the pitch angle ⁇ and azimuth angle of each point after regularization
  • the pitch angle ⁇ can be directly obtained from the vertical acquisition range of each laser scanner in the calibration file, while the azimuth angle need to pass the sampling interval Sure.
  • FIG. 4 is an expanded view of a cylindrical coordinate system provided by an embodiment of the present invention, in which the vertical interval and the horizontal interval can become concepts similar to the resolution in the image. Therefore, the vertical resolution theta and the horizontal resolution phi are respectively:
  • S3 Perform regularization preprocessing on the two-dimensional structure based on the geometric distortion measure to obtain a regularized structure.
  • step S2 After the vertical resolution and the horizontal resolution are determined in step S2, the radius r of each point from the center after regularization needs to be determined.
  • this embodiment uses the geometric distortion measurement idea to perform nearest neighbor interpolation to calculate the r component of the corresponding point after regularization processing, thereby ensuring that the geometric reconstruction quality is at D2 (point to plane)
  • the distortion can be controlled within a certain range.
  • the two-dimensional structure can be adjusted according to the point-to-surface geometric distortion measure (D2) to obtain a regularized structure.
  • D2 point-to-surface geometric distortion measure
  • FIG. 5 is a schematic diagram of point-to-surface interpolation processing provided by an embodiment of the present invention. Specifically,
  • the distance information is taken as the radius of the regularized current point from the center, that is, the r component in the cylindrical coordinates.
  • FIG. 6 is a comparison diagram before and after the regularization process provided by an embodiment of the present invention.
  • the regularized structure constructed by the above steps can ensure that the distortion from each point to the plane is zero, maintain the geometric structure information of the original point cloud, and have little impact on the performance of applications such as identification and automatic driving, and This regularized structure is extremely friendly to further processing of subsequent point clouds.
  • the two-dimensional structure can also be adjusted according to the point-to-point geometric distortion measure (D1) to obtain a regularized structure, which can not only ensure the geometric D1 distortion measure, but also ensure the point Model quality of the cloud.
  • D1 point-to-point geometric distortion measure
  • the two-dimensional structure can also be adjusted according to the comprehensive distortion measures of point-to-point (D1) and point-to-surface (D2) to obtain a regularized structure; in this way, the overall distortion measures of D1 and D2 of the geometry and D2 can be guaranteed at the same time.
  • Point cloud model quality
  • the two-dimensional structure can also be adjusted according to the point-to-line geometric distortion measure to obtain a regularized structure.
  • the geometric distortion measure based on statistical point-to-line is between the point-to-point and point-to-surface distortion measures, so that the overall distortion of the geometry D1 and D2 and the model quality of the point cloud can be guaranteed at the same time.
  • the point cloud data processing method provided by the present invention performs regularized preprocessing on the original input point cloud, so that the point cloud presents a regular distribution in both the horizontal and vertical directions, and increases the correlation between the data, so as to facilitate the subsequent point cloud processing.
  • the cloud data is further processed, and in the regularization process, the geometric distortion measure is used for the regularization process in the horizontal direction, which ensures the quality of the point cloud model.
  • FIG. 7 is a schematic diagram of a point cloud geometry encoding method provided by an embodiment of the present invention, including the following steps:
  • Step 1 Obtain raw point cloud data.
  • the original point cloud data is represented by Cartesian coordinates as (x, y, z).
  • Step 2 Perform regularization preprocessing on the original point cloud data to obtain a regularized structure.
  • the preprocessing method provided in the first embodiment above can be used to perform regularized preprocessing on the original point cloud data to obtain a regularized structure.
  • the cylindrical coordinates (x, y, z) are converted into a regular (r, j, i) structure.
  • Step 3 Determine the prediction mode of each point in the regularized structure, and use the selected prediction mode to perform geometric prediction on each point to obtain the information to be encoded.
  • a prediction tree structure is established based on the lidar calibration information.
  • the prediction mode of the current point is selected according to the prediction tree structure.
  • the established prediction tree is traversed in a depth-first order, and each node in the tree can only be predicted by its ancestors.
  • this embodiment sets the following four prediction modes:
  • p0, p1, p2 are the positions of the parent node, grandfather node, and great-grandfather node of the current node, respectively. According to the reconstruction quality, the best prediction mode can be selected for the current node for prediction.
  • geometric prediction is performed on each point in the prediction tree structure according to the selected prediction mode, and the geometric prediction residual of each point is obtained.
  • the geometric prediction includes cylindrical coordinate prediction and Cartesian coordinate prediction, and the specific process is as follows:
  • the prediction mode selects Mode0, that is, the cylindrical coordinates of the current node are not predicted, and the corresponding predicted cylindrical coordinates are (r min , j prev , i prev ), where r min is the entire
  • r min is the entire
  • the minimum value of the r component obtained after the coordinate transformation of the point cloud, if the current node has no parent node, then j prev and i prev are set to 0, otherwise it is the cylindrical coordinate component of the parent node.
  • the predicted value of Cartesian coordinates is obtained by inverse transformation of the cylindrical coordinates (r, j, i) of the point
  • the cylindrical coordinates of the current point can be predicted through the cylindrical coordinates of its parent node (r min , j prev , i prev ) to obtain the predicted value of the cylindrical coordinates of the current point ( r',j',i'), the Cartesian coordinate prediction value is obtained by inverse transformation of the original cylindrical coordinate (r,j,i)
  • the cylinder coordinates of the current point are predicted by the corresponding prediction methods, and the predicted value of the cylinder coordinates of the current point is (r',j',i' ), similarly, the Cartesian coordinate predicted value is obtained by the inverse transformation of the original cylindrical coordinate (r, j, i).
  • j prev represents the predicted azimuth of the current point
  • n represents the number of points that need to be skipped between the parent node and the current point according to the scanning speed, and if the Laser (laser scanner) of the current node is i, and the current node is adjacent to The Laser of i+1 has been encoded and decoded, then the parameter n can further use the corresponding position node n' of Laser as i+1 to perform differential prediction to obtain the prediction residual that needs to be skipped, namely:
  • the coding and decoding are performed sequentially according to each Laser, so it is necessary to temporarily store the node j component when the Laser has been coded to be i, for coding the Laser When i+1 ⁇ N, the prediction of the node j component at the corresponding position.
  • the prediction residuals in the cylindrical coordinate system (r r , r j , r i ), the prediction residuals in the Cartesian coordinate system (r x , r y , r z ), and the predictions that need to skip the number of points residual Together with other parameters that need to be encoded, such as the number of child nodes of the current node, the prediction mode of the current node and other information, one of them is used as the information to be encoded.
  • Step 4 Code the information to be coded in sequence to obtain a code stream of geometric information.
  • the number of child nodes of the current node needs to be encoded, and then the prediction mode of the current node, and the corresponding (r r , r j , r i ) and (r x , r y , r z ) prediction residuals and prediction residuals that need to skip the number of points
  • the point cloud geometric encoding method provided in this embodiment, by performing regular processing on the original input point cloud, makes the point cloud appear regular distribution in both the horizontal and vertical directions, increases the correlation between data, and improves the encoding efficiency At the same time, due to the regularization of the point cloud azimuth direction, it is not necessary to use an additional code stream to encode auxiliary information in the azimuth direction during encoding, which saves the code stream and improves the encoding efficiency.
  • the regularized structure when the regularized point is subjected to geometric coding, the regularized structure is effectively used, and the horizontal and vertical directions are combined with each other to perform predictive coding, thereby improving the geometric coding efficiency.
  • step 3 may also adopt the existing geometric coding mode based on prediction numbers to convert Cartesian coordinates (x, y, z) into cylindrical coordinates Then make a prediction and get the predicted value and and prediction residuals (r r , r ⁇ , r i ) and (r x , r y , r z ), and use lossless coding to
  • the components are encoded corresponding to the number of skipped points n, and the specific process is not described in detail here.
  • a mode switch can also be set to instruct whether to initialize the original point cloud data during the entire encoding process.
  • geom_enable_regular_flag is introduced to guide whether the regularization preprocessing scheme of the present invention is enabled for geometry in the entire encoding.
  • geom_enable_regular_flag is 1, it means open; otherwise, it is closed; see Appendix 1 for details.
  • the point cloud data preprocessing method provided in the first embodiment above, the point cloud data can be preprocessed first, and then the coding scheme provided in the second embodiment or the existing geometric prediction coding can be used. method to perform predictive coding on point cloud data to improve coding efficiency.
  • the prediction method provided in the second embodiment above can also be directly used to predict and encode the original point cloud data.
  • FIG. 8 is a schematic structural diagram of a point cloud geometry encoding device provided by an embodiment of the present invention, including:
  • the first data acquisition module 11 is used to acquire original point cloud data
  • the regularization module 12 performs regularization preprocessing on the original point cloud data to obtain a regularized structure
  • the first prediction module 13 is used to determine the prediction mode of each point in the regularized structure, and use the selected prediction mode to perform geometric prediction on each point to obtain the information to be encoded;
  • the encoding module 14 is used for sequentially encoding the information to be encoded to obtain a code stream of geometric information.
  • the apparatus provided in this embodiment can implement the encoding method provided in the second embodiment above, and the specific implementation process is not repeated here.
  • FIG. 9 is a schematic diagram of a point cloud geometry decoding method provided by an embodiment of the present invention, including:
  • Step 1 Acquire the geometric information code stream, and perform decoding to obtain decoded data; wherein the decoded data includes the prediction mode of the current node.
  • Step 2 Perform geometric prediction on the current node according to the prediction mode to obtain prediction residuals; wherein the prediction residuals include prediction residuals in a cylindrical coordinate system and prediction residuals in a Cartesian coordinate system.
  • the prediction mode selects Mode0, that is, the cylindrical coordinates of the current node are not predicted, and the corresponding predicted cylindrical coordinates are (r min , j prev , i prev ), where r min is the entire
  • r min is the entire
  • j prev and i prev are set to 0, otherwise it is the cylindrical coordinate component of the parent node.
  • the predicted value of Cartesian coordinates is obtained by inverse transformation of the original cylindrical coordinates (r, j, i) of the point
  • the cylindrical coordinates of the current point can be predicted through the cylindrical coordinates of its parent node (r min , j prev , i prev ) to obtain the predicted value of the cylindrical coordinates of the current point ( r',j',i'), the Cartesian coordinate prediction value is obtained by inverse transformation of the original cylindrical coordinate (r,j,i)
  • the cylinder coordinates of the current point are predicted by the corresponding prediction methods, and the predicted value of the cylinder coordinates of the current point is (r',j',i' ), similarly, the Cartesian coordinate predicted value is obtained by the inverse transformation of the original cylindrical coordinate (r, j, i).
  • the cylindrical coordinates (r, j, i) of the current point are predicted, and the corresponding prediction residuals (r r , r j , r i ) in the cylindrical coordinate system are obtained.
  • j prev represents the predicted azimuth of the current point
  • n represents the number of points to be skipped between the parent node and the current point according to the scanning speed. Note that if the laser of the current node is i, and the adjacent Laser of the current node is i-1 and the encoding and decoding has been completed, then the parameter n is recovered by using the corresponding position node n' whose Laser is i-1, namely:
  • Step 3 Rebuild the prediction tree structure according to the prediction residuals in the cylindrical coordinate system, and perform coordinate transformation on the points in the prediction tree structure to obtain the predicted Cartesian coordinates of the current point.
  • the position of the current point in the prediction tree can be further determined according to the reconstructed cylindrical coordinates (r, j, i), thereby reconstructing the prediction tree.
  • i is the LaserID corresponding to the point, and the prior information of each Laser is different, that is, the elevation angle ⁇ and the height zLaser in the vertical direction are different, so the elevation angle corresponding to the i-th Laser is ⁇ (i), the height in the vertical direction is zLaser(i).
  • Step 4 Reconstruct the point cloud according to the prediction residual and the predicted Cartesian coordinates in the Cartesian coordinate system, and obtain the reconstructed point cloud data.
  • FIG. 10 is a schematic structural diagram of a point cloud geometry decoding apparatus provided by an embodiment of the present invention, including:
  • the second data acquisition module 21 is used to acquire the geometric information code stream, and decode it to obtain decoded data; wherein, the decoded data includes the prediction mode of the current node;
  • the second prediction module 22 is configured to perform geometric prediction on the current node according to the prediction mode to obtain prediction residuals; wherein the prediction residuals include prediction residuals in the cylindrical coordinate system and prediction residuals in the Cartesian coordinate system;
  • the prediction tree reconstruction module 23 is used to reconstruct the prediction tree structure according to the prediction residuals under the cylindrical coordinate system, and perform coordinate conversion on the points in the prediction tree structure to obtain the predicted Cartesian coordinates of the current point;
  • the point cloud reconstruction module 24 reconstructs the geometric point cloud according to the geometric prediction value and the prediction residual of the current node, and obtains the reconstructed point cloud data.
  • the apparatus provided in this embodiment can implement the decoding method provided in the fourth embodiment above, and the specific implementation process is not repeated here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Geometry (AREA)
  • Computer Graphics (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)
  • Compression Of Band Width Or Redundancy In Fax (AREA)

Abstract

本发明公开了一种点云数据的预处理方法及点云几何编解码方法、装置，所述预处理方法包括：对原始点云数据进行坐标转换，得到原始点云在柱面坐标系下的表示；将柱面坐标系展开，得到二维结构；基于几何失真测度对二维结构进行规则化处理，得到规则化结构。所述编码方法包括：对经过预处理的原始点云数据进行预测编码，得到几何信息码流。本发明提供的点云几何编码方法，通过对原始点云进行规则化预处理，使得点云在水平和竖直方向均呈现规则化分布，更好的体现了点云的空间相关性，提高了编码效率。

Description

点云数据的预处理方法及点云几何编解码方法、装置
本申请要求于2021年02月08日提交中国专利局、申请号为202110180985.0、申请名称为“点云数据的预处理方法及点云几何编解码方法、装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明属于点云数据处理技术领域，具体涉及一种点云数据的预处理方法及点云几何编解码方法、装置。
背景技术
在点云G-PCC(Geometry-based Point Cloud Compression，基于几何的点云压缩编码)编码器框架中，点云的几何信息和每点所对应的属性信息是分开进行编码的。目前G-PCC的几何编解码可分为基于八叉树的几何编解码和基于预测树几何编解码。基于预测树的几何编码首先对输入点云进行排序，并在编码端通过利用两种不同的方式建立预测树结构。然后基于预测树的结构，遍历预测树中的每个节点，通过选取不同的预测模式对节点的几何位置信息进行预测得到预测残差，并且利用量化参数对几何预测残差进行量化。最终通过不断迭代，对预测树节点位置信息的预测残差、预测树结构以及量化参数等进行编码，生成二进制码流。
基于激光雷达标定信息的预测树编码是目前常用的几何编码方式。对于激光雷达的每一个激光扫描器而言，隶属于同一个激光扫描器的采集点在圆柱坐标系下应是规则分布的。然而，由于噪声、测量误差和设备抖动等因素导致实际数据呈现非均匀分布，从而导致数据之间相关性较差，预测精度和编码效率较低。
然而，基于预测树的点云编解码技术仅通过利用激光雷达设备的部分参数来建立树结构，该树结构并未充分体现点云的空间相关性，不利于点云的 预测及熵编码，从而影响了编码效率。而现有的G-PCC方法仅通过垂直方向校正确定每点与激光扫描器之间的关系，造成在编码时需要引入其他的变量辅助水平方向信息的编码，从而增加了需要编码的信息量，降低了几何编码效率。
发明内容
为了解决现有技术中存在的上述问题，本发明提供了一种基于规则化结构的点云几何预测编解码方法及装置。本发明要解决的技术问题通过以下技术方案实现：
一种点云数据预处理方法，包括：
对原始点云数据进行坐标转换，得到原始点云在柱面坐标系下的表示；
将所述柱面坐标系展开，得到二维结构；
基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构。
在本发明的一个实施例中，基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构，包括：
根据点到面的几何失真测度对所述二维结构进行调整，得到规则化结构。
在本发明的一个实施例中，根据点到面的几何失真测度对所述二维结构进行调整，得到规则化结构，包括：
查找所述二维结构中方位角和俯仰角方向上距离当前点最近的点；
通过该距离当前点最近的点的角度信息构造由原点发出的射线；
根据所述当前点及其法线构建平面；
求取射线与平面的交点，并记录从原点到该交点位置的距离；
将该距离作为规则化后的当前点距离中心的半径；
重复上述步骤，完成所有点的处理，得到原始点云数据的规则化结构。
在本发明的一个实施例中，基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构，还包括：
根据点到点的几何失真测度对所述二维结构进行调整，以得到规则化结 构；或者
根据点到点和点到面的综合失真测度对所述二维结构进行调整，以得到规则化结构；或者
根据点到线的几何失真测度对所述二维结构进行调整，以得到规则化结构。
本发明的另一个实施例还提供了一种点云几何编码方法，其特征在于，包括：
获取原始点云数据；
采用上述实施例所述的预处理方法对所述原始点云数据进行规则化预处理，得到规则化结构；
确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息；
对所述待编码信息依次进行编码，得到几何信息码流。
在本发明的一个实施例中，确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息，包括：
基于激光雷达标定信息建立预测树结构；
根据所述预测树结构选择每个点的预测模式；
根据选取的预测模式对所述预测树结构中的每个点进行几何预测，得到每个点的几何预测残差；
将所述几何预测残差作为部分待编码信息。
在本发明的一个实施例中，根据选取的预测模式对所述预测树结构中的每个点进行几何预测，得到每个点的几何预测残差，包括：
根据当前节点类型和所选的预测模式，对当前节点的柱面坐标(r,j,i)进行预测，得到当前节点柱面坐标系下的预测值(r',j',i')及预测残差(r r,r j,r i)；其中，当前点的方位角的预测值j'按以下公式进行计算：
j'＝j prev+n；
其中，j prev表示当前点的预测方位角；n表示父节点与当前点之间按照扫描速度需要跳过的点数，其预测残差
Figure PCTCN2022075379-appb-000001
为
Figure PCTCN2022075379-appb-000002
n'表示与当前点相邻的且已编码的节点需要跳过的点数；
根据当前点的笛卡尔坐标(x,y,z)与预测笛卡尔坐标
Figure PCTCN2022075379-appb-000003
进行差分预测得到笛卡尔坐标系下的预测残差(r x,r y,r z)。
本发明的另一个实施例还提供了一种点云几何编码装置，包括：
第一数据获取模块，用于获取原始点云数据；
规则化模块，对所述原始点云数据进行规则化预处理，得到规则化结构；
第一预测模块，用于确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息；
编码模块，用于对所述待编码信息依次进行编码，得到几何信息码流。
本发明的又一个实施例还提供了一种点云几何解码方法，包括：
获取几何信息码流，并进行解码得到解码数据；其中，所述解码数据包括当前节点的预测模式；
根据所述预测模式对当前节点进行几何预测，得到预测残差；其中，所述预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差；
根据所述柱面坐标系下的预测残差重建预测树结构，并对所述预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标；
根据所述笛卡尔坐标系下的预测残差和所述预测笛卡尔坐标进行点云重建，得到重建点云数据。
本发明的再一个实施例还提供了一种点云几何解码装置，包括：
第二数据获取模块，用于获取几何信息码流，并进行解码得到解码数据；其中，所述解码数据包括当前节点的预测模式；
第二预测模块，用于根据所述预测模式对当前节点进行几何预测，得到预测残差；其中，所述预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差；
预测树重建模块，用于根据所述柱面坐标系下的预测残差重建预测树结构，并对所述预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标；
点云重建模块，根据所述当前节点的几何预测值和预测残差重建几何点云，得到重建点云数据。
本发明的有益效果：
1、本发明提供的点云数据处理方法通过对原始输入点云进行了规则化预处理，使得点云在水平和竖直方向均呈现规则化分布，更好的体现了点云的空间相关性，以便于后续进行点云数据的进一步处理；
2、本发明在进行规则化处理时，采用几何失真测度对二维结构进行调整，保证了点云模型质量；
3、本发明提供的点云几何编码方法在对点云数据进行预处理时，对点云方位角方向上进行了规则化处理，使得在进行编码时，无需使用额外码流编码方位角方向上的辅助信息，节省了码流，提升了编码效率；
4、本发明对经过规则化处理之后的点云进行几何编码时，通过有效的利用规则化结构，将水平和垂直方向相互结合进行预测编码，提升了几何编码效率。
以下将结合附图及实施例对本发明做进一步详细说明。
附图说明
图1是本发明实施例提供的一种点云数据预处理方法示意图；
图2是本发明实施例提供的激光雷达结构示意图；
图3是本发明实施例提供的原始采集数据分布结构示意图；
图4是本发明实施例提供的柱面坐标系展开图；
图5是本发明实施例提供的点到面的插值处理示意图
图6是本发明实施例提供的规则化处理的前后对比图；
图7是本发明实施例提供的一种点云几何编码方法示意图；
图8是本发明实施例提供的一种点云几何编码装置结构示意图；
图9是本发明实施例提供的一种点云几何解码方法示意图；
图10是本发明实施例提供的一种点云几何解码装置结构示意图。
具体实施方式
下面结合具体实施例对本发明做进一步详细的描述，但本发明的实施方式不限于此。
实施例一
请参见图1，图1是本发明实施例提供的一种点云数据预处理方法示意图，包括：
S1：对原始点云数据进行坐标转换，得到原始点云在柱面坐标系下的表示。
具体地，原始点云数据通常由一组三维空间点组成，每个空间点都记录了自身的几何位置信息，以及颜色、反射率、法线等额外的属性信息。其中，点云的几何位置信息一般是基于笛卡尔坐标系进行表示的，即利用点的x，y，z坐标进行表示。原始点云数据可通过3D扫描设备例如激光雷达等获取，也可通过各种平台提供的公共数据集获得。在本实施例中，设获取到的原始点云数据的几何位置信息基于笛卡尔坐标系表示为(x,y,z)。需要说明的是，原始点云数据的几何位置信息的表示方法不限于笛卡尔坐标。
具体地，在对原始点云进行坐标转换之前，还可以对其进行行量化处理和重排序处理，以方便后续进行预测编码。
请参见图2，图2是本发明实施例提供的激光雷达结构示意图；激光雷达由多个激光扫描器构成，在描述一个激光雷达时，所谓的“线数”即其中所包含的激光扫描器个数，这些激光扫描器沿激光雷达的中心轴两侧分布，且具有不同的俯仰角θ i，从而可以获取空间中垂直方向上不同物体的空间信息。每个都可以看作一个相对独立的采集系统。采集时通过底座的旋转，所有的激光扫描器按照一定的采样率获取空间中的物体的位置信息。
在本实施例中，可根据现有的转换公式进将原始点云数据的笛卡尔坐标(x,y,z)转换为柱面坐标(r,φ,i)，得到原始点云在柱面坐标系下的表示。
请参见图3，图3是本发明实施例提供的原始采集数据在柱面坐标系下的分布结构示意图。在理想情况下，由激光雷达获取的点云数据在柱面坐标系中呈现出沿方位角和俯仰角方向的均匀分布。然而由于噪声、机械抖动以及激光雷达内部坐标系对齐等因素，使点云呈现出非均匀分布的特点。
S2：将柱面坐标系展开，得到二维结构。
在本实施例中，首先利用激光雷达的结构、采集参数确定了规则化后每一点的俯仰角θ和方位角
Figure PCTCN2022075379-appb-000004
此处，俯仰角θ可以直接从标定文件中每一个激光扫描器的垂直采集范围获得，而方位角
Figure PCTCN2022075379-appb-000005
需通过采样间隔
Figure PCTCN2022075379-appb-000006
确定。
具体地，请参见图4，图4是本发明实施例提供的柱面坐标系展开图，其中，垂直方向的间隔和水平方向间隔则可以变成类似于图像中分辨率的概念。因此垂直方向分辨率theta和水平方向分辨率phi分别为：
theta＝laserNum；
Figure PCTCN2022075379-appb-000007
S3：基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构。
在步骤S2确定了垂直方向分辨率和水平方向分辨率之后，还需要确定规则化后每一点距离中心的半径r。
由于半径的选择直接决定了规则化后的点云是否可以维持与输入点云相同的几何结构。而对于识别、自动驾驶之类的应用而言，几何结构的失真直接决定了此类应用的性能。因此，为了尽可能地减少模型的几何结构失真，本实施例通过利用几何失真测度思想进行最近邻插值计算经过规则化处理之后对应点的r分量，进而保证几何重建质量在D2(点到平面)的失真可以控制在一定范围。
在本实施例中，可根据点到面的几何失真测度(D2)对二维结构进行调整，得到规则化结构。
请参见图5，图5是本发明实施例提供的点到面的插值处理示意图，具体地，
1、先查找二维结构中方位角和俯仰角方向上距离当前点最近的点；
2、通过该点的角度信息构造由原点(origin)发出的射线；
3、根据当前点p i及其法线构建平面；
4、求射线与平面的交点，并记录从原点到此该交点位置的距离。
5、将该距离信息作为规则化后的当前点距离中心的半径，也即柱面坐标中的r分量。
6、重复上述步骤，完成所有点的处理，得到原始点云数据的规则化结构。
至此，完成了将点云的柱面坐标(r,φ,i)的规则化预处理，得到了规则化处理后的坐标(r,j,i)。请参见图6，图6是本发明实施例提供的规则化处理的前后对比图。
本实施例通过上述步骤构造的规则化结构，可保证每个点到平面的失真为零，维持了原始点云所具有的几何结构信息，对识别、自动驾驶等应用的性能影响较小，且此种规则化结构对于后续点云的进一步处理极为友善。
在本发明的另一个实施例中，还可根据点到点的几何失真测度(D1)对二维结构进行调整，以得到规则化结构，这样既可以保证几何的D1失真测度，又能够保证点云的模型质量。
此外，还可以根据点到点(D1)和点到面(D2)的综合失真测度对二维结构的进行调整，以得到规则化结构；这样同时可以保证几何的D1和D2的总体失真测度和点云模型质量。
进一步地，还可以根据点到线的几何失真测度对二维结构进行调整，以得到规则化结构。基于统计点到线的几何失真测度处于点到点和点到面的失真测度之间，这样就也可以同时保证几何D1和D2的总体失真以及点云的模 型质量。
本发明提供的点云数据处理方法通过对原始输入点云进行了规则化预处理，使得点云在水平和竖直方向均呈现规则化分布，增加了数据之间相关性，以便于后续进行点云数据的进一步处理，且在进行规则化处理时，采用几何失真测度进行水平方向的规则化处理，保证了点云模型质量。
实施例二
请参见图7，图7是本发明实施例提供的一种点云几何编码方法示意图，包括以下步骤：
步骤1：获取原始点云数据。
在本实施例中，原始点云数据采用笛卡尔坐标表示为(x,y,z)。
步骤2：对原始点云数据进行规则化预处理，得到规则化结构。
具体地，可采用上述实施例一提供的预处理方法对原始点云数据进行规则化预处理，得到规则化结构。
更具体地，通过规则化处理之后，将柱面坐标(x,y,z)转换成了规则的的(r,j,i)结构。
步骤3：确定规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息。
首先，基于激光雷达标定信息建立预测树结构。
然后，根据预测树结构选择当前点的预测模式。
在本实施例中，采取深度优先顺序遍历建立的预测树，树中的每个节点只能由其祖先来预测。
进一步地，本实施例设置以下四种预测模式：
Mode0:无预测
Mode1:Delta预测(即p0)
Mode2:Linear预测(即2p0-p1)
Mode3:Parallelogram预测(即p0+p1-p2)
其中，p0、p1、p2分别为当前节点的父结点、祖父结点、曾祖父结点的位置。可根据重建质量对当前节点选取最佳的预测模式进行预测。
接着，根据选取的预测模式对预测树结构中的每个点进行几何预测，得到每个点的几何预测残差。在本实施例中，几何预测包括柱面坐标预测和笛卡尔坐标预测，具体过程如下：
根据当前节点类型和所选的预测模式，对当前节点的柱面坐标(r,j,i)进行预测，得到当前节点柱面坐标系下的预测值(r',j',i')及预测残差(r r,r j,r i)；
若当前节点为预测树的根节点，则预测模式选择Mode0，即当前节点的柱面坐标无预测，对应的柱面坐标预测值为(r min,j prev,i prev),其中r min为整个点云进行坐标转换之后得到的r分量最小值，如果当前节点没有父节点，那么j prev,i prev置为0，否则为父节点的柱面坐标分量。笛卡尔坐标预测值为点的柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000008
若当前节点不是根节点，且预测模式选择Mode1时，当前点的柱面坐标通过其父节点的柱面坐标(r min,j prev,i prev)进行预测可得当前点柱面坐标预测值(r',j',i')，笛卡尔坐标预测值为原始柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000009
若当前节点不是根节点，且预测模式选择Mode2或Mode3时，当前点的柱面坐标通过对应的预测方式进行预测，可得当前点柱面坐标的预测值为(r',j',i')，同样，笛卡尔坐标预测值为原始柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000010
利用当前节点最佳的预测模式，对当前点的柱面坐标(r,j,i)进行预测，得到圆柱坐标系下的对应的预测残差(r r,r j,r i)。
需要说明的是，当前点的方位角的预测值j'按以下公式进行计算：
j'＝j prev+n；
其中，j prev表示当前点的预测方位角；n表示父节点与当前点之间按照扫描速度需要跳过的点数，并且如果当前节点的Laser(激光扫描器)为i，且当前 节点的相邻的Laser为i+1已经完成编解码，那么参数n可以进一步利用Laser为i+1的相应位置节点n'进行差分预测，得到需要跳过点数的预测残差，即：
Figure PCTCN2022075379-appb-000011
此外，由于在本实施例是基于规则化结构进行几何预测编码时,按照每个Laser依次进行编解码的，因此需要对已经编码完成Laser为i时的节点j分量进行临时存储，用于编码Laser为i+1～N时，对相应位置的节点j分量的预测。
用当前点的笛卡尔坐标(x,y,z)与预测笛卡尔坐标
Figure PCTCN2022075379-appb-000012
进行差分预测得到笛卡尔坐标系下的预测残差(r x,r y,r z)。
最后，将柱面坐标系下的预测残差(r r,r j,r i)、笛卡尔坐标系下的预测残差(r x,r y,r z)、以及需要跳过点数的预测残差
Figure PCTCN2022075379-appb-000013
连同其余需要编码的参数，如当前节点的子节点数、当前节点的预测模式等信息其一作为待编码信息。
步骤4：对待编码信息依次进行编码，得到几何信息码流。
具体地，对于每个节点的待编码信息，首先，需要对当前节点的子节点数目进行编码,其次编码当前节点的预测模式,以及当前节点分别对应的(r r,r j,r i)和(r x,r y,r z)预测残差以及需要跳过点数的预测残差
Figure PCTCN2022075379-appb-000014
至此，完成点云的几何预测编码。
本实施例提供的点云几何编码方法，通过对原始输入点云进行了规则化处理，使得点云在水平和竖直方向均呈现规则化分布，增加了数据之间相关性，提高了编码效率；同时，由于对点云方位角方向上的规则化处理，使得在进行编码时，无需使用额外码流编码方位角方向上的辅助信息，节省了码流，提升了编码效率。
本实施例对经过规则化处理之后的点进行几何编码时，通过有效的利用规则化结构，水平和垂直方向相互结合进行预测编码，提升了几何的编码效率。
在本发明的另一个实施例中，步骤3还可以采用现有的基于预测数的几 何编码模式，将笛卡尔坐标(x,y,z)转换成柱面坐标
Figure PCTCN2022075379-appb-000015
然后进行预测，得到预测值
Figure PCTCN2022075379-appb-000016
和
Figure PCTCN2022075379-appb-000017
以及预测残差(r r,r φ,r i)和(r x,r y,r z)，并采用无损编码方式对
Figure PCTCN2022075379-appb-000018
分量对应跳过的点数n进行编码，具体过程在此不做详细说明。
实施例三
在上述实施例二的基础上，还可以设置一个模式开关，以指导整个编码过程中，是否对原始点云数据进行初始化。
具体地，在gps(Geometry parameter set syntax)参数集引入：geom_enable_regular_flag,用以指导在整个编码中，几何是否开启本发明的规则化预处理方案。geom_enable_regular_flag为1时，表示开启；否则关闭；具体的见附表1。
当启用规则化预处理方案时，可以在上述实施例一提供的点云数据预处理方法，先对点云数据进行预处理，然后再采用实施例二提供的编码方案或者现有的几何预测编码方法，对点云数据进行预测编码，以提高编码效率。
当不启用规则化预处理方案时，也可以直接采用上述实施例二提供的预测方法对原始点云数据进行预测并编码。
实施例四
在上述实施例二的基础上，本实施例提供了一种点云几何编码装置。请参见图8，图8是本发明实施例提供的一种点云几何编码装置结构示意图，包括：
第一数据获取模块11，用于获取原始点云数据；
规则化模块12，对原始点云数据进行规则化预处理，得到规则化结构；
第一预测模块13，用于确定规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息；
编码模块14，用于对待编码信息依次进行编码，得到几何信息码流。
本实施例提供的装置可实现上述实施例二提供的编码方法，具体实现过程在此不再赘述。
实施例五
请参见图9，图9是本发明实施例提供的一种点云几何解码方法示意图，包括：
步骤一：获取几何信息码流，并进行解码得到解码数据；其中，解码数据包括当前节点的预测模式。
步骤二：根据预测模式对当前节点进行几何预测，得到预测残差；其中，预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差。
根据解码得到的当前点采用的预测模式及其祖先的重建柱面坐标对当前点的柱面坐标进行预测并生成对应的预测值，具体过程如下：
若当前节点为预测树的根节点，则预测模式选择Mode0，即当前节点的柱面坐标无预测，对应的柱面坐标预测值为(r min,j prev,i prev),其中r min为整个点云进行坐标转换之后得到的r分量最小值，如果当前节点没有父节点，那么j prev,i prev置为0，否则为父节点的柱面坐标分量。笛卡尔坐标预测值为点的原始柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000019
若当前节点不是根节点，且预测模式选择Mode1时，当前点的柱面坐标通过其父节点的柱面坐标(r min,j prev,i prev)进行预测可得当前点柱面坐标预测值(r',j',i')，笛卡尔坐标预测值为原始柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000020
若当前节点不是根节点，且预测模式选择Mode2或Mode3时，当前点的柱面坐标通过对应的预测方式进行预测，可得当前点柱面坐标的预测值为(r',j',i')，同样，笛卡尔坐标预测值为原始柱面坐标(r,j,i)逆转换得到的
Figure PCTCN2022075379-appb-000021
利用当前节点最佳的预测模式,对当前点的柱面坐标(r,j,i)进行预测,得到圆柱坐标系下的对应的预测残差(r r,r j,r i)。
同编码端一样，需要说明的是，当前点的方位角的预测值j'按以下公式进行计算：
j'＝j prev+n；
其中，j prev表示当前点的预测方位角；n表示父节点与当前点之间按照扫描速度需要跳过的点数。注意，如果当前节点的laser为i，并且当前节点的相邻的Laser为i-1已经完成编解码，那么参数n利用Laser为i-1的相应位置节点n'进行恢复得到，即：
Figure PCTCN2022075379-appb-000022
至此，得到柱面坐标系下的预测残差、需要跳过点数以及笛卡尔坐标系下的预测残差。
步骤三：根据柱面坐标系下的预测残差重建预测树结构，并对预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标。
具体地，利用解码得到的柱面坐标残差(r r,r j,r i)和当前点的预测柱面坐标(r',j',i')计算当前点的重建柱面坐标(r,j,i)。
(r,j,i)＝(r',j',i')+(r r,r j,r i)。
得到当前点的重建柱面坐标后就可以进一步根据该重建柱面坐标(r,j,i)确定当前点在预测树中的位置，从而重建预测树。
将当前点的重建柱面坐标(r,j,i)按照以下公式转换为笛卡尔坐标
Figure PCTCN2022075379-appb-000023
Figure PCTCN2022075379-appb-000024
即为当前点的预测笛卡尔坐标。
Figure PCTCN2022075379-appb-000025
Figure PCTCN2022075379-appb-000026
Figure PCTCN2022075379-appb-000027
Figure PCTCN2022075379-appb-000028
其中，i为点对应的LaserID，每个Laser的先验信息不同即仰角θ和在垂直方向上的高度zLaser不同，因此第i个Laser对应的仰角为θ(i)，在垂直方向上的高度为zLaser(i)。
步骤四：根据笛卡尔坐标系下的预测残差和预测笛卡尔坐标进行点云重建，得到重建点云数据。
按照以下公式利用解码得到的笛卡尔坐标残差(r x,r y,r z)和当前点的预测笛卡尔坐标
Figure PCTCN2022075379-appb-000029
计算当前点的重建笛卡尔坐标(x,y,z)。
Figure PCTCN2022075379-appb-000030
至此，完成点云的解码，得到了重建的体素化后的点云。
实施例六
在上述实施例五的基础上，本实施例提供了一种点云几何解码装置。请参见图10，图10是本发明实施例提供的一种点云几何解码装置结构示意图，包括：
第二数据获取模块21，用于获取几何信息码流，并进行解码得到解码数据；其中，解码数据包括当前节点的预测模式；
第二预测模块22，用于根据预测模式对当前节点进行几何预测，得到预测残差；其中，预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差；
预测树重建模块23，用于根据柱面坐标系下的预测残差重建预测树结构，并对预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标；
点云重建模块24，根据当前节点的几何预测值和预测残差重建几何点云，得到重建点云数据。
本实施例提供的装置可实现上述实施例四提供的解码方法，具体实现过程在此不再赘述。
以上内容是结合具体的优选实施方式对本发明所作的进一步详细说明，不能认定本发明的具体实施只局限于这些说明。对于本发明所属技术领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干简单推演或替换，都应当视为属于本发明的保护范围。
附表1：Geometry parameter set syntax
Figure PCTCN2022075379-appb-000031
Figure PCTCN2022075379-appb-000032
Figure PCTCN2022075379-appb-000033

Claims (10)

  1. 一种点云数据预处理方法，其特征在于，包括：
    对原始点云数据进行坐标转换，得到原始点云在柱面坐标系下的表示；
    将所述柱面坐标系展开，得到二维结构；
    基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构。
  2. 根据权利要求1所述的点云数据预处理方法，其特征在于，基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构，包括：
    根据点到面的几何失真测度对所述二维结构进行调整，得到规则化结构。
  3. 根据权利要求2所述的点云数据预处理方法，其特征在于，根据点到面的几何失真测度对所述二维结构进行调整，得到规则化结构，包括：
    查找所述二维结构中方位角和俯仰角方向上距离当前点最近的点；
    通过该距离当前点最近的点的角度信息构造由原点发出的射线；
    根据所述当前点及其法线构建平面；
    求取射线与平面的交点，并记录从原点到该交点位置的距离；
    将该距离作为规则化后的当前点距离中心的半径；
    重复上述步骤，完成所有点的处理，得到原始点云数据的规则化结构。
  4. 根据权利要求1所述的点云数据预处理方法，其特征在于，基于几何失真测度对所述二维结构进行规则化预处理，得到规则化结构，还包括：
    根据点到点的几何失真测度对所述二维结构进行调整，以得到规则化结构；或者
    根据点到点和点到面的综合失真测度对所述二维结构进行调整，以得到规则化结构；或者
    根据点到线的几何失真测度对所述二维结构进行调整，以得到规则化结构。
  5. 一种点云几何编码方法，其特征在于，包括：
    获取原始点云数据；
    采用如权利要求1-4任一项所述的预处理方法对所述原始点云数据进行规则化预处理，得到规则化结构；
    确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息；
    对所述待编码信息依次进行编码，得到几何信息码流。
  6. 根据权利要求5所述的点云几何编码方法，其特征在于，确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息，包括：
    基于激光雷达标定信息建立预测树结构；
    根据所述预测树结构选择每个点的预测模式；
    根据选取的预测模式对所述预测树结构中的每个点进行几何预测，得到每个点的几何预测残差；
    将所述几何预测残差作为部分待编码信息。
  7. 根据权利要求6所述的点云几何编码方法，其特征在于，根据选取的预测模式对所述预测树结构中的每个点进行几何预测，得到每个点的几何预测残差，包括：
    根据当前节点类型和所选的预测模式，对当前节点的柱面坐标(r,j,i)进行预测，得到当前节点柱面坐标系下的预测值(r',j',i')及预测残差(r r,r j,r i)；其中，当前点的方位角的预测值j'按以下公式进行计算：
    j'＝j prev+n；
    其中，j prev表示当前点的预测方位角；n表示父节点与当前点之间按照扫描速度需要跳过的点数，其预测残差
    Figure PCTCN2022075379-appb-100001
    为
    Figure PCTCN2022075379-appb-100002
    n'表示与当前点相邻的且已编码的节点需要跳过的点数；
    根据当前点的笛卡尔坐标(x,y,z)与预测笛卡尔坐标
    Figure PCTCN2022075379-appb-100003
    进行差分预测得到笛卡尔坐标系下的预测残差(r x,r y,r z)。
  8. 一种点云几何编码装置，其特征在于，包括：
    第一数据获取模块(11)，用于获取原始点云数据；
    规则化模块(12)，对所述原始点云数据进行规则化预处理，得到规则化结构；
    第一预测模块(13)，用于确定所述规则化结构中每个点的预测模式，并利用所选的预测模式对每个点进行几何预测，得到待编码信息；
    编码模块(14)，用于对所述待编码信息依次进行编码，得到几何信息码流。
  9. 一种点云几何解码方法，其特征在于，包括：
    获取几何信息码流，并进行解码得到解码数据；其中，所述解码数据包括当前节点的预测模式；
    根据所述预测模式对当前节点进行几何预测，得到预测残差；其中，所述预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差；
    根据所述柱面坐标系下的预测残差重建预测树结构，并对所述预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标；
    根据所述笛卡尔坐标系下的预测残差和所述预测笛卡尔坐标进行点云重建，得到重建点云数据。
  10. 一种点云几何解码装置，其特征在于，包括：
    第二数据获取模块(21)，用于获取几何信息码流，并进行解码得到解码数据；其中，所述解码数据包括当前节点的预测模式；
    第二预测模块(22)，用于根据所述预测模式对当前节点进行几何预测，得到预测残差；其中，所述预测残差包括柱面坐标系下的预测残差以及笛卡尔坐标系下的预测残差；
    预测树重建模块(23)，用于根据所述柱面坐标系下的预测残差重建预测树结构，并对所述预测树结构中的点进行坐标转换，得到当前点的预测笛卡尔坐标；
    点云重建模块(24)，根据所述当前节点的几何预测值和预测残差重建几何 点云，得到重建点云数据。
PCT/CN2022/075379 2021-02-08 2022-02-07 点云数据的预处理方法及点云几何编解码方法、装置 WO2022166957A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP22749236.0A EP4246442A4 (en) 2021-02-08 2022-02-07 METHOD FOR PREPROCESSING POINT CLOUD DATA, METHOD AND DEVICE FOR CODING THE POINT CLOUD GEOMETRY AND METHOD AND DEVICE FOR DECODING THE POINT CLOUD GEOMETRY
US18/267,977 US20240062429A1 (en) 2021-02-08 2022-02-07 Point cloud data preprocessing method, point cloud geometry coding method and apparatus, and point cloud geometry decoding method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110180985.0A CN114913252A (zh) 2021-02-08 2021-02-08 点云数据的预处理方法及点云几何编解码方法、装置
CN202110180985.0 2021-02-08

Publications (2)

Publication Number Publication Date
WO2022166957A1 true WO2022166957A1 (zh) 2022-08-11
WO2022166957A9 WO2022166957A9 (zh) 2023-05-19

Family

ID=82742007

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/075379 WO2022166957A1 (zh) 2021-02-08 2022-02-07 点云数据的预处理方法及点云几何编解码方法、装置

Country Status (4)

Country Link
US (1) US20240062429A1 (zh)
EP (1) EP4246442A4 (zh)
CN (1) CN114913252A (zh)
WO (1) WO2022166957A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11875541B2 (en) 2020-10-07 2024-01-16 Qualcomm Incorporated Predictive geometry coding in G-PCC
WO2024060161A1 (zh) * 2022-09-22 2024-03-28 上海交通大学 编解码方法、编码器、解码器以及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024065408A1 (zh) * 2022-09-29 2024-04-04 Oppo广东移动通信有限公司 编解码方法、码流、编码器、解码器以及存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108632607A (zh) * 2018-05-09 2018-10-09 北京大学深圳研究生院 一种基于多角度自适应帧内预测的点云属性压缩方法
CN109214982A (zh) * 2018-09-11 2019-01-15 大连理工大学 一种基于双圆柱投影模型的三维点云成像方法
WO2020189943A1 (ko) * 2019-03-15 2020-09-24 엘지전자 주식회사 포인트 클라우드 데이터 송신 장치, 포인트 클라우드 데이터 송신 방법, 포인트 클라우드 데이터 수신 장치 및 포인트 클라우드 데이터 수신 방법

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108632607A (zh) * 2018-05-09 2018-10-09 北京大学深圳研究生院 一种基于多角度自适应帧内预测的点云属性压缩方法
CN109214982A (zh) * 2018-09-11 2019-01-15 大连理工大学 一种基于双圆柱投影模型的三维点云成像方法
WO2020189943A1 (ko) * 2019-03-15 2020-09-24 엘지전자 주식회사 포인트 클라우드 데이터 송신 장치, 포인트 클라우드 데이터 송신 방법, 포인트 클라우드 데이터 수신 장치 및 포인트 클라우드 데이터 수신 방법

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11875541B2 (en) 2020-10-07 2024-01-16 Qualcomm Incorporated Predictive geometry coding in G-PCC
US11935270B2 (en) 2020-10-07 2024-03-19 Qualcomm Incorporated Predictive geometry coding in G-PCC
WO2024060161A1 (zh) * 2022-09-22 2024-03-28 上海交通大学 编解码方法、编码器、解码器以及存储介质

Also Published As

Publication number Publication date
EP4246442A1 (en) 2023-09-20
WO2022166957A9 (zh) 2023-05-19
CN114913252A (zh) 2022-08-16
US20240062429A1 (en) 2024-02-22
EP4246442A4 (en) 2024-06-05

Similar Documents

Publication Publication Date Title
WO2022166957A1 (zh) 点云数据的预处理方法及点云几何编解码方法、装置
US6438266B1 (en) Encoding images of 3-D objects with improved rendering time and transmission processes
EP4052474A1 (en) Angular priors for improved prediction in tree-based point cloud coding
US20230290007A1 (en) Point Cloud Encoding and Decoding Method and Device Based on Two-Dimensional Regularization Plane Projection
WO2022166961A1 (zh) 面向大规模点云的二维规则化平面投影及编解码方法
WO2022166968A1 (zh) 基于二维规则化平面投影的点云编解码方法及装置
WO2022166967A1 (zh) 基于二维规则化平面投影的点云编解码方法及装置
WO2022166969A1 (zh) 基于二维规则化平面投影的点云序列编解码方法及装置
CN113763539B (zh) 一种基于图像和三维输入的隐式函数三维重建方法
Beemelmanns et al. 3D Point Cloud Compression with Recurrent Neural Network and Image Compression Methods
CN115393452A (zh) 一种基于非对称自编码器结构的点云几何压缩方法
WO2022166966A1 (zh) 基于二维规则化平面投影的点云编解码方法及装置
WO2022247705A1 (zh) 一种点云属性信息的预测编解码方法及装置
CN115412717B (zh) 一种点云方位角信息的预测编解码方法及装置
US20240119636A1 (en) Point cloud encoding and decoding method and device based on two-dimensional regularization plane projection
WO2024074122A1 (en) Method, apparatus, and medium for point cloud coding
WO2024083194A1 (en) Method, apparatus, and medium for point cloud coding
WO2024074121A1 (en) Method, apparatus, and medium for point cloud coding
WO2024086123A1 (en) Decoding method, encoding method, decoder and encoder
Zhang et al. The 3D model compression method based on sparse representation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22749236

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18267977

Country of ref document: US

ENP Entry into the national phase

Ref document number: 2022749236

Country of ref document: EP

Effective date: 20230615

NENP Non-entry into the national phase

Ref country code: DE