Distance segmented sampling method and system for LiDAR point cloud compression

Abstract

The invention discloses a distance segmented sampling method and system for LiDAR point cloud compression, and the method comprises the steps: carrying out the down-sampling and up-sampling at two ends of a codec, so as to reduce the redundancy of point cloud data, and achieve the efficient point cloud reconstruction. In the down-sampling process, the three-dimensional coordinates of the LiDAR point cloud are converted into a polar coordinate system, segmented scale transformation is carried out on the polar coordinate system, and the dimensionality of point cloud data is reduced. And performing neighbor search on the source point cloud through a KD-Tree algorithm to obtain point cloud data after down-sampling. In the up-sampling process, an occupation graph is reconstructed through the decoded point cloud, width expansion and splicing are carried out, and finally up-sampled point cloud data are generated. The method has the advantages that the over-sampling characteristic of the near data of the LiDAR point cloud data is utilized, and the to-be-coded data volume is reduced through down-sampling, so that the compression efficiency of the LiDAR point cloud data is improved.

Description

Technical Field

[0001] The present invention relates to the field of LiDAR data processing technology, and in particular to a distance segmented sampling method and system for LiDAR point cloud compression. Background Art

[0002] With the rapid development of LiDAR technology, LiDAR point cloud data has been widely used in many fields such as unmanned driving, robot navigation, environmental monitoring, urban modeling and geographic information system (GIS). LiDAR sensors can accurately obtain the three-dimensional spatial information of target objects and generate rich point cloud data. However, with the rapid growth of point cloud data, the problems of storing, transmitting and processing these huge data sets have gradually become a bottleneck restricting its widespread application.

[0003] In traditional LiDAR data processing, the size and complexity of point cloud data place high demands on computing resources, storage space, and transmission bandwidth. In order to reduce these burdens, researchers have proposed a variety of point cloud compression techniques. These methods usually simplify the point cloud data representation, reduce data redundancy, and achieve efficient storage and transmission. However, existing point cloud compression methods usually have the following problems:

[0004] 1. Data distortion problem: Existing compression methods usually adopt global sampling or coarse grid division methods, which may lead to the loss of detail information during the compression process, thus affecting the reconstruction effect, especially in application scenarios with high precision requirements.

[0005] 2. Low compression efficiency: Traditional point cloud compression algorithms fail to fully utilize the local characteristics of LiDAR point cloud data, such as the density, spatial distribution and object structure characteristics of data points, resulting in low algorithm efficiency when performing large-scale data compression.

[0006] 3. High consumption of transmission and storage resources: In some real-time applications, the processing of LiDAR point cloud data needs to consider the limitations of transmission bandwidth and storage space. Existing methods often cannot effectively solve the storage and real-time transmission problems under large amounts of data.

[0007] 4. Lack of flexibility and adaptability: Many traditional point cloud sampling methods need to be manually adjusted according to different point cloud data characteristics in specific applications, and lack the ability to automatically adapt to different point cloud densities and distributions, resulting in poor versatility of the algorithm.

[0008] Therefore, how to achieve efficient compression, transmission and reconstruction while ensuring the quality of point cloud data, especially in a dynamically changing environment, has become a technical problem that needs to be solved urgently. Summary of the invention

[0009] In view of the defects of the prior art, the present invention provides a distance segmented sampling method and system for LiDAR point cloud compression. Through refined occupancy map processing, downsampling and upsampling mechanisms, not only can the LiDAR point cloud data be effectively compressed, but also a high accuracy can be maintained during the reconstruction process.

[0010] In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0011] A distance segmented sampling method for LiDAR point cloud compression, comprising:

[0012] a) Downsampling and upsampling are performed at both ends of the codec;

[0013] b) Use P O , P D , P R , P U They represent the source point cloud, the downsampled point cloud, the decoded reconstructed point cloud and the upsampled point cloud respectively. O ,M D Indicates the original size (width W s ) and the occupancy map after distance segment scale transformation (the width of the i-th segment is W i express);

[0014] c) The downsampling process includes:

[0015] i. Convert the (x, y, z) coordinates of the point cloud into polar coordinates And generate the occupancy map group M with θ and r as horizontal and vertical coordinates O , whose abscissa is azimuth, and ordinate is distance, and each scan line corresponds to an occupancy map;

[0016] ii. According to the distance r of the LiDAR point cloud, the data is divided into N segments, and each segment is scaled in the azimuth dimension to obtain the occupancy map M after segment scale transformation D ;

[0017] iii. M D From polar coordinates Convert back to (x, y, z) coordinates to get M D The corresponding point cloud P1;

[0018] iv. Find each point in P1 in the source point cloud P O The nearest point in the image is obtained by downsampling the point cloud P. D ;

[0019] d) The upsampling process includes:

[0020] i. The decoded point cloud P RThe (x, y, z) coordinates of the point are converted to polar coordinates The occupancy map M is generated with θ and r as the horizontal and vertical coordinates. D ;

[0021] ii. Expand the width of each sub-segment from W i Change back to W s , and concatenate the sub-segments into the final upsampled occupancy map M in a center-aligned manner O ;

[0022] iii. Insubordination Convert back to (x, y, z) coordinates to obtain point cloud P2;

[0023] iv. Find the decoded point cloud P in the upsampled point cloud P2 R The nearest neighbor of each point in P2 is R The most recent use of P R The points in are replaced, and the other points remain unchanged to obtain the final upsampled point cloud P U .

[0024] Furthermore, according to the distance information of the LiDAR point cloud, the data is divided into N segments, and each segment is scaled to obtain an occupancy map after segmented scale transformation, as follows:

[0025] According to the LiDAR point cloud, R min To R max The data within meters is divided into N segments, represented by subscripts 0 to N-1 from near to far, and each segment is scaled to change the horizontal resolution from the width W of the original occupancy map to s Down to the final width W of the i-th subsegment i , the width of each occupancy map is W = [W0, W1, ..., W i ,…,W N-1 ] indicates that the i-th segment occupancy map data is scaled to obtain the occupancy map M after segment scale transformation D ; where N is the number of segments, W i is the width of each segment during downsampling.

[0026] Furthermore, the downsampling process further includes: setting the value of each pixel of the occupancy map to 1 or 0, indicating whether there is point cloud data at that location; for each segment, adjusting according to the distance range and scale to ensure the uniformity and compression efficiency of the point cloud data; the width and height of the occupancy map are not affected by data sampling or object distribution, and are only determined by sampling resolution and scale transformation.

[0027] Furthermore, the upsampling process further includes:

[0028] For each pixel with a value of 1 in the downsampled occupancy map, restore it to (x, y, z) three-dimensional data;

[0029] Point cloud P R For each point in the point cloud P, use the KD-Tree algorithm O Find its nearest point and set P O Replace the points in with the corresponding points in P.

[0030] As a preference, W i The calculation formula is as follows:

[0031]

[0032] Where, ΔW i is the width adjustment amount of the i-th sub-segment, which is used to make slight adjustments to the width of each sub-segment to make it conform to the predetermined structure; N is the number of segments.

[0033] The present invention also discloses a compression-oriented LiDAR sampling system, which can be used to implement the above-mentioned distance segmented sampling method for LiDAR point cloud compression, and specifically includes the following modules:

[0034] The point cloud preprocessing module is used to receive the original LiDAR point cloud data and convert the three-dimensional coordinates into a polar coordinate system to generate an occupancy map.

[0035] The downsampling module is used to perform downsampling processing on the input LiDAR point cloud data. The downsampling module includes:

[0036] Coordinate conversion unit, converting the three-dimensional coordinates (x, y, z) of the point cloud into a polar coordinate system And generate the occupancy map group M with θ and r as horizontal and vertical coordinates O , where the abscissa represents the azimuth, the ordinate represents the distance, and each scan line corresponds to an occupancy map;

[0037] The segmentation processing unit divides the data into N segments according to the distance information r of the LiDAR point cloud, and scales each segment in the azimuth dimension to obtain the occupancy map M after segment scale transformation D ;

[0038] The coordinate transformation unit transforms the occupancy map M after the segment scale transformation D From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) and get M D The corresponding point cloud P1;

[0039] KD-Tree search unit, using KD-Tree algorithm, finds each point in P1 in the source point cloud P OThe nearest point in the image is obtained by downsampling the point cloud P. D .

[0040] The encoder module is used to encode the downsampled point cloud P D Encoding is performed to compress point cloud data into a smaller representation format for easy transmission or storage;

[0041] The decoder module is used to receive the encoded point cloud data and decode it into the reconstructed point cloud P R .

[0042] Upsampling module, used to decode the point cloud P R Upsampling is performed to restore the spatial distribution of the original point cloud data. The upsampling module includes:

[0043] The coordinate conversion unit converts the decoded point cloud P R The coordinates (x, y, z) are converted to polar coordinates The occupancy map M is generated with θ and r as the horizontal and vertical coordinates. D ;

[0044] The segment expansion unit divides the decoded occupancy map into N segments, and the width of each segment is consistent with the corresponding width during downsampling; the width of each sub-segment is expanded from W i Change back to W s , and concatenate the sub-segments into the final upsampled occupancy map M in a center-aligned manner O ;

[0045] The coordinate conversion unit converts the upsampled occupancy map M O From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) to obtain point cloud P2;

[0046] KD-Tree search unit, using the KD-Tree algorithm, finds the decoded point cloud P in the upsampled point cloud P2 R The nearest neighbor of each point in P2 is R The nearest point replaces P R The corresponding points in , the rest of the points remain unchanged, and the final up-sampled point cloud P is obtained U ;

[0047] Point cloud post-processing module: used to process the upsampled point cloud P U For further processing, such as data visualization, storage or transmission.

[0048] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned distance segmented sampling method for LiDAR point cloud compression is implemented.

[0049] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the distance segmented sampling method for LiDAR point cloud compression is implemented.

[0050] Compared with the prior art, the advantages of the present invention are:

[0051] Improve data compression efficiency: By downsampling and upsampling the occupancy map of LiDAR point cloud data, the present invention can effectively compress the volume of the original point cloud data and reduce the cost of storage and transmission. Through scale transformation and fine sub-segment width adjustment, the compression effect can be further optimized while ensuring data quality.

[0052] Reduce information loss: Using the KD-Tree algorithm to perform nearest neighbor search on point cloud data can maximize the retention of key spatial information in the downsampling process, thereby avoiding the large information loss problem common in traditional compression methods and maintaining high reconstruction accuracy.

[0053] Adapt to different application requirements: Through user-defined downsampling rate S and number of segments N, the system can flexibly adjust the compression level and detail retention to meet the needs of point cloud data processing in different application scenarios. Whether it is fast processing in a low-resolution environment or detailed analysis under high-precision requirements, it can meet different needs.

[0054] Efficient upsampling and reconstruction capabilities: The upsampling process uses appropriate scale transformation and width expansion to efficiently restore the decoded point cloud data to the approximate state of the original point cloud. At the same time, it effectively avoids the distortion problem introduced by errors in the upsampling process and ensures high-quality reconstruction of the point cloud.

[0055] Widely applicable scalability: The LiDAR point cloud sampling method of the present invention is not only applicable to existing LiDAR systems, but also compatible with a variety of codecs, with good scalability and adaptability. By adjusting the parameters in the algorithm, the system can run stably in different hardware and software environments.

[0056] Performance improvement: The present invention has achieved significant performance improvement on the KITTI dataset (including Residential, Campus, Road and City subsets), and has improved the D1-PSNR metric distortion by 36.32% compared to the traditional method. This effect proves that the present invention can significantly improve the processing efficiency and quality of point cloud data in practical applications.

[0057] Optimize the point cloud data processing flow: By combining downsampling and upsampling techniques, especially by refining the occupancy map, the present invention can balance accuracy and efficiency during data compression, transmission and reconstruction, optimize the overall processing flow of point cloud data, and is suitable for different LiDAR point cloud data application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of LiDAR point cloud encoding and decoding according to an embodiment of the present invention;

[0059] Figure 2 is a flow chart of LiDAR point cloud downsampling according to an embodiment of the present invention;

[0060] Figure 3 This is a flow chart of LiDAR point cloud upsampling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0062] The present invention provides a distance segmented sampling method for LiDAR point cloud compression, such as Figure 1 As shown, downsampling and upsampling are performed at both ends of the codec.

[0063] This results in a version with fewer points during actual encoding, which can reduce the actual number of code streams. At the decoding end, upsampling is used to complete the point cloud data, which can effectively improve the efficiency of point cloud compression encoding. Figure 1 P O ,P D ,P R ,P U They represent the source point cloud, the downsampled point cloud, the decoded and reconstructed point cloud, and the upsampled point cloud respectively; S is the downsampling rate of the point cloud origin input by the user, and its size affects the overall sampling ratio; N is the N segments into which the point cloud is divided, which can be defined by the user.

[0064] 1. Downsampling process

[0065] The downsampling process is a preprocessing operation before the compression of LiDAR point cloud in this invention. The main purpose is to reduce the amount of point cloud data by performing segmented scale transformation on the original LiDAR point cloud data. The process includes the following steps:

[0066] (1.1) Coordinate transformation:

[0067] Convert the original point cloud data in the (x, y, z) three-dimensional coordinate system to the polar coordinate system In this step, r represents the distance between the point cloud data and the sensor, θ and Represents the angles of the point cloud in the horizontal and vertical directions respectively.

[0068] Use θ and r as the horizontal and vertical coordinates to generate the occupancy map M O For each scan line of data, a binary occupancy map is generated based on whether there is an occupied position in the point cloud. If there is point cloud data at the current position, the position is assigned a value of 1; otherwise, it is assigned a value of 0. For a 64-line LiDAR sensor, this process will generate 64 occupancy maps, also called occupancy map groups.

[0069] (1.2) Segmented scaling:

[0070] According to the spatial characteristics of LiDAR point cloud data (i.e., point clouds at close distances are denser and point clouds at long distances are sparser), point cloud data are grouped within a certain distance range (e.g., R min =5 to R max =30 meters) is divided into N (N=5) segments.

[0071] Each occupancy map is resized to reduce the width of each segment and shorten the actual number of points. N-1 ] indicates that the width W of the original occupancy map s Gradually reduce to the width W of each segment i， W s >=W i In this way, the point cloud data at a close distance retains a higher resolution, while the point cloud data at a long distance reduces the resolution to achieve a compression effect. The width of each scaled occupancy map W i Represented by array W.

[0072] W i One of the calculation formulas is as follows:

[0073]

[0074] Among them, S and N are parameters entered by the user, representing the number of segments and the maximum number of sampling scales.

[0075] (1.3) The occupancy map is converted back to three-dimensional coordinates:

[0076] After the scale transformation, the occupancy map of each segment is converted back to the (x, y, z) coordinates in the three-dimensional space.

[0077] In this process, due to the downsampling operation, the number of point clouds will be significantly reduced, and the number of points in the obtained three-dimensional point cloud data will be less than that of the original data.

[0078] (1.4) Use KD-Tree to find the nearest point:

[0079] In order to ensure the accuracy of the data, the KD-Tree algorithm is used to O Find the nearest point of each remaining point and use these points as the downsampled point cloud data P D Through the nearest neighbor search of KD-Tree, it is ensured that the downsampled point cloud data represents the structure of the original point cloud as much as possible.

[0080] 2. Coding process

[0081] Downsampled point cloud data P D The point cloud is encoded by the encoder to obtain a compressed code stream. During the encoding process, different LiDAR point cloud encoding technologies can be flexibly selected to compress the point cloud to obtain a code stream s. The compression parameters can be adjusted according to actual needs.

[0082] 3. Decoding process

[0083] Downsampled point cloud data P D The point cloud data P is reconstructed by the decoder. R .

[0084] 4. Upsampling Process

[0085] The upsampling process is a post-decoding processing step, the purpose of which is to reconstruct a point cloud data as close as possible to the original point cloud by restoring and supplementing the point cloud data lost during downsampling. The process includes the following steps:

[0086] (4.1) Coordinate transformation and occupancy map drawing:

[0087] First, the decoded point cloud data P R The (x, y, z) coordinates of the point are converted to polar coordinates Where r represents the distance from the point cloud data to the origin, θ and Represents the angle of the point cloud in the horizontal and vertical directions. Then, the occupancy map M is drawn with θ and r as the horizontal and vertical coordinates. D .

[0088] Since each point cloud data is composed of 64 groups of sub-point cloud data obtained by 64 radar scanning lines, each group of sub-point cloud data corresponds to an occupancy map. That is, one point cloud data will generate 64 occupancy maps through conversion.

[0089] (4.2) Scale transformation and occupancy map recovery:

[0090] For each sub-segment of each occupancy map, the inverse scale transformation is performed according to the parameters in the downsampling process, and its width is changed from W to i Expand to the width W of the original point cloud data occupancy map sAt this point, the width of all sub-segments will be the same, which is W s . Because W s Greater than W i , so this step actually increases the number of points in the point cloud and achieves upsampling

[0091] (4.3) Restoring three-dimensional data and KD-Tree nearest neighbor search:

[0092] For each pixel point with a pixel value of 1 in the obtained upsampled occupancy map, it is restored to three-dimensional coordinate data to obtain a point cloud P. Each pixel point is mapped to the (x, y, z) coordinates of the three-dimensional space, thereby restoring the upsampled point cloud data.

[0093] In order to improve the accuracy of upsampling, the KD-Tree algorithm (or other algorithms) can be used to decode the point cloud P R Find the nearest neighbor point of each point cloud. Specifically, for P R For each point in P, use the KD-Tree search algorithm to find its nearest point in point cloud P2 and use P R Substitute the corresponding points in P2 with the points in P2, and keep the other points unchanged to obtain P U .

[0094] V. Experiment and Effect

[0095] By testing on the KITTI dataset (including four subsets: Residential, Campus, Road, and City), the distance segmented sampling method for LiDAR point cloud compression of the present invention can effectively improve the efficiency of LiDAR point cloud data encoding and decoding. The experimental results show that after adopting the method of the present invention, the compression effect has a performance improvement of 36.32% when the D1-PSNR metric distortion is used.

[0096] VI. Conclusion

[0097] The present invention effectively improves the compression efficiency of LiDAR point cloud data by performing coordinate transformation, segmented scale transformation, downsampling and upsampling on LiDAR point cloud data. This method can not only significantly reduce the burden of data storage and transmission, but also maintain the structure and distribution of point cloud data to ensure the quality of the decoded point cloud. Through reasonable downsampling rate and segmentation settings, users can flexibly adjust the compression ratio according to specific needs to achieve the purpose of optimizing the encoding effect.

[0098] The present invention is used for compression, storage and reconstruction of large-scale point cloud data and has broad application prospects, especially in the fields of unmanned driving, robot navigation and three-dimensional modeling.

[0099] In another embodiment of the present invention, a distance segmented sampling system for LiDAR point cloud compression is provided. The system can be used to implement the above-mentioned distance segmented sampling method for LiDAR point cloud compression, and specifically includes the following modules:

[0100] The point cloud preprocessing module is used to receive the original LiDAR point cloud data and convert the three-dimensional coordinates into a polar coordinate system to generate an occupancy map.

[0101] The downsampling module is used to perform downsampling processing on the input LiDAR point cloud data. The downsampling module includes:

[0102] Coordinate conversion unit, converting the three-dimensional coordinates (x, y, z) of the point cloud into a polar coordinate system And generate the occupancy map group M with θ and r as horizontal and vertical coordinates O , where the abscissa represents the azimuth, the ordinate represents the distance, and each scan line corresponds to an occupancy map;

[0103] The segmentation processing unit divides the data into N segments according to the distance information r of the LiDAR point cloud, and scales each segment in the azimuth dimension to obtain the occupancy map M after segment scale transformation D ;

[0104] The coordinate transformation unit transforms the occupancy map M after the segment scale transformation D From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) and get M D The corresponding point cloud P1;

[0105] KD-Tree search unit, using KD-Tree algorithm, finds the nearest point of each point in P1 in the source point cloud P, and obtains the downsampled point cloud P D .

[0106] The encoder module is used to encode the downsampled point cloud P D Encoding is performed to compress point cloud data into a smaller representation format for easy transmission or storage;

[0107] The decoder module is used to receive the encoded point cloud data and decode it into the reconstructed point cloud P R .

[0108] Upsampling module, used to decode the point cloud P R Upsampling is performed to restore the spatial distribution of the original point cloud data. The upsampling module includes:

[0109] The coordinate conversion unit converts the decoded point cloud P R The coordinates (x, y, z) are converted to polar coordinates The occupancy map M is generated with θ and r as the horizontal and vertical coordinates.D ;

[0110] The segment expansion unit divides the decoded occupancy map into N segments, and the width of each segment is consistent with the corresponding width during downsampling; the width of each sub-segment is expanded from W i Change back to W s , and concatenate the sub-segments into the final upsampled occupancy map M in a center-aligned manner O ;

[0111] The coordinate conversion unit converts the upsampled occupancy map M O From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) to obtain point cloud P2;

[0112] KD-Tree search unit, using the KD-Tree algorithm, finds the decoded point cloud P in the upsampled point cloud P2 R The nearest neighbor of each point in P is replaced by the point in P2. R The corresponding points in , the rest of the points remain unchanged, and the final up-sampled point cloud P is obtained U ;

[0113] Point cloud post-processing module: used to process the upsampled point cloud P U For further processing, such as data visualization, storage or transmission.

[0114] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the distance segmented sampling method for LiDAR point cloud compression.

[0115] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0116] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the distance segmented sampling method for LiDAR point cloud compression in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor.

[0117] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A distance segmented sampling method for LiDAR point cloud compression, characterized in that: include: a) Downsampling and upsampling are performed at both ends of the codec; b) Use P O , P D , P R , P U They represent the source point cloud, the downsampled point cloud, the decoded reconstructed point cloud and the upsampled point cloud respectively. O ,M D Represents the occupancy map of the original size and the occupancy map after the distance segment scale transformation; c) The downsampling process includes: i. Convert the (x, y, z) coordinates of the point cloud into polar coordinates And generate the occupancy map group M with θ and r as horizontal and vertical coordinates O , whose abscissa is azimuth, and ordinate is distance, and each scan line corresponds to an occupancy map; ii. According to the distance r of the LiDAR point cloud, the data is divided into N segments, and each segment is scaled in the azimuth dimension to obtain the occupancy map M after segment scale transformation D ; iii. M D From polar coordinates Convert back to (x, y, z) coordinates to get M D The corresponding point cloud P1; iv. Find each point in P1 in the source point cloud P O The nearest point in the image is obtained by downsampling the point cloud P. D ; d) The upsampling process includes: i. The decoded point cloud P R The (x, y, z) coordinates of the point are converted to polar coordinates The occupancy map M is generated with θ and r as the horizontal and vertical coordinates. D ; ii. Expand the width of each sub-segment from W i Change back to W s , and concatenate the sub-segments into the final upsampled occupancy map M in a center-aligned manner O ; iii. Insubordination Convert back to (x, y, z) coordinates to obtain point cloud P2; iv. Find the decoded point cloud P in the upsampled point cloud P2 R The nearest neighbor of each point in P2 is R The most recent use of P R The points in are replaced, and the other points remain unchanged to obtain the final upsampled point cloud P U .

2. The distance segmented sampling method for LiDAR point cloud compression according to claim 1, characterized in that: According to the distance information of the LiDAR point cloud, the data is divided into N segments, and each segment is scaled to obtain an occupancy map after segmented scale transformation, as follows: According to the LiDAR point cloud, R min To R max The data within meters is divided into N segments, represented by subscripts 0 to N-1 from near to far, and each segment is scaled to change the horizontal resolution from the width W of the original occupancy map to s Down to the final width W of the i-th subsegment i , the width of each occupancy map is W = [W0, W1, ..., W i ,…,W N-1 ] indicates that the i-th segment occupancy map data is scaled to obtain the occupancy map M after segment scale transformation D ; where N is the number of segments, W i is the width of each segment during downsampling.

3. The distance segmented sampling method for LiDAR point cloud compression according to claim 1, characterized in that: The downsampling process further includes: setting the value of each pixel of the occupancy map to 1 or 0, indicating whether there is point cloud data at that location; for each segment, adjusting according to the distance range and scale to ensure the uniformity and compression efficiency of the point cloud data; the width and height of the occupancy map are not affected by data sampling or object distribution, and are only determined by sampling resolution and scale transformation.

4. The distance segmented sampling method for LiDAR point cloud compression according to claim 1, characterized in that: The upsampling process further comprises: For each pixel with a value of 1 in the downsampled occupancy map, restore it to (x, y, z) three-dimensional data; Point cloud P R For each point in the point cloud P, use the KD-Tree algorithm O Find its nearest point and set P O Replace the points in with the corresponding points in P.

5. The distance segmented sampling method for LiDAR point cloud compression according to claim 1, characterized in that: W i The calculation formula is as follows: Where, ΔW i is the width adjustment amount of the i-th sub-segment, which is used to make slight adjustments to the width of each sub-segment to make it conform to the predetermined structure; N is the number of segments.

6. A distance segmented sampling system for LiDAR point cloud compression, characterized by: The system can be used to implement the distance segmented sampling method for LiDAR point cloud compression according to any one of claims 1 to 5, and specifically includes the following modules: Point cloud preprocessing module, which is used to receive the original LiDAR point cloud data and convert the three-dimensional coordinates into polar coordinate system to generate occupancy map; The downsampling module is used to perform downsampling processing on the input LiDAR point cloud data. The downsampling module includes: Coordinate conversion unit, converting the three-dimensional coordinates (x, y, z) of the point cloud into a polar coordinate system And generate the occupancy map group M with θ and r as horizontal and vertical coordinates O , where the abscissa represents the azimuth, the ordinate represents the distance, and each scan line corresponds to an occupancy map; The segmentation processing unit divides the data into N segments according to the distance information r of the LiDAR point cloud, and scales each segment in the azimuth dimension to obtain the occupancy map M after segment scale transformation D ; The coordinate transformation unit transforms the occupancy map M after the segment scale transformation D From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) and get M D The corresponding point cloud P1; KD-Tree search unit, using KD-Tree algorithm, finds each point in P1 in the source point cloud P O The nearest point in the image is obtained by downsampling the point cloud P. D ; The encoder module is used to encode the downsampled point cloud P D Encoding is performed to compress point cloud data into a smaller representation format for easy transmission or storage; The decoder module is used to receive the encoded point cloud data and decode it into the reconstructed point cloud P R ; Upsampling module, used to decode the point cloud P R Upsampling is performed to restore the spatial distribution of the original point cloud data. The upsampling module includes: The coordinate conversion unit converts the decoded point cloud P R The coordinates (x, y, z) are converted to polar coordinates The occupancy map M is generated with θ and r as the horizontal and vertical coordinates. D ; The segment expansion unit divides the decoded occupancy map into N segments, and the width of each segment is consistent with the corresponding width during downsampling; the width of each sub-segment is expanded from W i Change back to W s , and concatenate the sub-segments into the final upsampled occupancy map M in a center-aligned manner O ; The coordinate conversion unit converts the upsampled occupancy map M O From polar coordinates Convert back to the three-dimensional coordinate system (x, y, z) to obtain point cloud P2; KD-Tree search unit, using the KD-Tree algorithm, finds the decoded point cloud P in the upsampled point cloud P2 R The nearest neighbor of each point in P2 is R The nearest point replaces P R The corresponding points in , and the other points remain unchanged, to obtain the final upsampled point cloud; Point cloud post-processing module: used to process the upsampled point cloud P U For further processing, such as data visualization, storage or transmission.

7. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the distance segmented sampling method for LiDAR point cloud compression according to one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the distance segmented sampling method for LiDAR point cloud compression as described in one of claims 1 to 5 is implemented.

Images