Point cloud data processing method, point cloud data processing system, and computer storage medium

Abstract

The present application relates to the technical field of data processing, and more particularly to a point cloud data processing method, a point cloud data processing system and a computer storage medium. The point cloud data processing method according to one aspect of the present application comprises: performing rasterization processing on a point cloud space of point cloud data to form a plurality of three-dimensional point cloud grids; selectively entering a sparse operation mode based on point cloud distribution of the point cloud space of the point cloud data and the plurality of three-dimensional point cloud grids; and performing the following operations in the sparse operation mode: taking a point cloud grid with point cloud data in the plurality of three-dimensional point cloud grids as a target point cloud grid and generating a grid mapping relationship based on a corresponding storage address of each target point cloud grid, and performing a three-dimensional convolution operation on each target point cloud grid based on the grid mapping relationship to generate an output point cloud grid.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically to point cloud data processing methods, point cloud data processing systems, and computer storage media. Background Technology

[0002] With the development of artificial intelligence and computer vision technologies, and the increasing emphasis on driving safety in vehicle-assisted driving and autonomous driving, technologies related to autonomous driving have become a current research hotspot. From a sensor perspective, LiDAR (Light Detection and Ranging) provides high-precision sensing data while being less affected by the environment, thus gradually becoming an indispensable sensor in the field of autonomous driving. Since laser light only produces an echo when it illuminates an object, the raw point cloud data acquired by LiDAR is typically distributed in a discrete and non-uniform manner, resulting in a large and sparse point cloud space. Usually, feature extraction is required from the raw point cloud data for target detection tasks.

[0003] Currently, in point cloud-based object detection methods, the unstructured point cloud space is typically divided into voxel spaces, and features are extracted from different voxel spaces by directly calculating 3D convolutions, ultimately used for point cloud data perception tasks. However, since raw point cloud data usually presents as a huge and sparse point cloud space, directly calculating 3D convolutions leads to a large amount of redundant computation and places high demands on memory and bandwidth. Once limited by global memory bandwidth, the utilization rate of computational resources will drop sharply. Summary of the Invention

[0004] To address or at least alleviate one or more of the above problems, the following technical solutions are provided.

[0005] According to a first aspect of this application, a point cloud data processing method is provided, the method comprising the following steps: rasterizing the point cloud space of the point cloud data to form a plurality of three-dimensional point cloud gratings; selectively entering a sparse operation mode based on the point cloud distribution of the point cloud space of the point cloud data and the plurality of three-dimensional point cloud gratings; and performing the following operations in the sparse operation mode: taking the point cloud gratings containing point cloud data among the plurality of three-dimensional point cloud gratings as target point cloud gratings and generating a raster mapping relationship based on the corresponding storage address of each target point cloud grating; and performing a three-dimensional convolution operation on each target point cloud grating based on the raster mapping relationship to generate an output point cloud grating.

[0006] According to an embodiment of this application, a point cloud data processing method is described, wherein rasterizing the point cloud space of point cloud data to form multiple three-dimensional point cloud grids includes: determining the size information of the three-dimensional point cloud grids using a calibration dataset, wherein the calibration dataset has the same probability distribution as the point cloud data; and rasterizing the point cloud space of the point cloud data based on the size information to form the multiple three-dimensional point cloud grids.

[0007] According to one embodiment or any of the above embodiments of the point cloud data processing method, generating a raster mapping relationship based on the corresponding storage address of each target point cloud raster includes: determining the size information of a mapping matrix based on the size information of the point cloud space and the size information of the three-dimensional point cloud raster; creating a mapping matrix based on the size information of the mapping matrix; and using the mapping matrix to establish a correspondence between each target point cloud raster and its corresponding storage address to generate the raster mapping relationship.

[0008] The point cloud data processing method according to one embodiment or any of the above embodiments of this application further includes: determining whether the size of the output point cloud grid is consistent with that of the target point cloud grid; and rearranging the output point cloud grid in response to determining that the size of the output point cloud grid is inconsistent with that of the target point cloud grid, so that the size of the output point cloud grid is consistent with that of the target point cloud grid.

[0009] The point cloud data processing method according to one embodiment or any of the above embodiments of this application further includes: filtering the output point cloud raster using a mask, wherein the elements in the mask are determined based on the dimension of the feature vector of the point cloud data contained in the target point cloud raster.

[0010] According to the point cloud data processing method of one embodiment or any of the above embodiments of this application, the elements in the mask are determined by: setting the elements in the mask to zero when the dimension of the feature vector of the point cloud data contained in the target point cloud raster is zero; and setting the elements in the mask to one when the dimension of the feature vector of the point cloud data contained in the target point cloud raster is not zero.

[0011] According to one embodiment or any of the above embodiments of the point cloud data processing method, selectively entering a sparse operation mode based on the point cloud distribution in the point cloud space of the point cloud data and the plurality of three-dimensional point cloud grids includes: inputting the point cloud distribution in the point cloud space of the point cloud data and the plurality of three-dimensional point cloud grids into a sparse operation evaluation model, and determining whether to enter the sparse operation mode based on the output result of the sparse operation evaluation model.

[0012] The point cloud data processing method according to one embodiment or any of the above embodiments of this application further includes performing the following operations in response to selecting not to enter the sparse operation mode: dividing the point cloud space of the point cloud data into voxel spaces; generating voxel feature maps by sampling the features of the point cloud data in each voxel space; and performing three-dimensional convolution or sparse convolution operations on the voxel feature maps.

[0013] According to a second aspect of this application, a point cloud data processing system is provided, the system comprising: a memory; a processor coupled to the memory; and a computer program stored on the memory and running on the processor, the execution of the computer program causing the execution of the steps of the point cloud data processing method according to a first aspect of this application.

[0014] According to a third aspect of this application, a computer storage medium is provided, comprising instructions that, when executed, perform the steps of the point cloud data processing method according to a first aspect of this application.

[0015] The point cloud data processing scheme according to one or more embodiments of this application, by using point cloud rasters containing point cloud data as target point cloud rasters and performing 3D convolution operations on each target point cloud raster, can transform sparse vector operations into dense matrix operations for point cloud data, thereby saving memory, reducing computational load, shortening computation time, and improving hardware utilization. The point cloud data processing scheme according to one or more embodiments of this application can be applied to various scenarios such as intelligent robot interaction, autonomous driving, and assisted driving, thereby achieving fast and accurate 3D target detection. Attached Figure Description

[0016] The above and / or other aspects and advantages of this application will become clearer and more readily understood from the following description taken in conjunction with the accompanying drawings, in which the same or similar elements are denoted by the same reference numerals. The drawings include:

[0017] Figure 1 A flowchart of a point cloud data processing method according to one or more embodiments of this application is shown.

[0018] Figure 2 A schematic diagram illustrating the generation of raster mapping relationships according to one or more embodiments of this application is shown.

[0019] Figure 3 A schematic diagram of point cloud data processing according to one or more embodiments of this application is shown.

[0020] Figure 4 A block diagram of a point cloud data processing system according to one or more embodiments of this application is shown. Detailed Implementation

[0021] The present application will now be described more fully with reference to the accompanying drawings, which illustrate exemplary embodiments thereof. However, the present application may be implemented in various forms and should not be construed as being limited to the embodiments given herein. The foregoing embodiments are intended to make the disclosure herein complete and thorough, so as to more fully convey the scope of protection of the present application to those skilled in the art.

[0022] In this specification, terms such as “comprising” and “including” indicate that, in addition to having the units and steps that are directly and explicitly stated in the specification and claims, the technical solution of this application does not exclude the presence of other units and steps that are not directly or explicitly stated.

[0023] Unless otherwise specified, terms such as “first” and “second” do not indicate the order of units in terms of time, space, size, etc., but are merely used to distinguish between units.

[0024] In the following, various exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings.

[0025] Figure 1 A flowchart of a point cloud data processing method according to one or more embodiments of this application is shown.

[0026] like Figure 1 As shown, in step S101, the point cloud space of the point cloud data is rasterized to form multiple three-dimensional point cloud grids.

[0027] Optionally, point cloud data refers to three-dimensional data determined based on the echo signals reflected back to the vehicle from environmental points in the vehicle's driving environment after receiving electromagnetic waves sent to it by the vehicle. This three-dimensional data includes the coordinates of the environmental points in a three-dimensional coordinate system. Point cloud space refers to the three-dimensional space formed by the three-dimensional point cloud data. In one or more embodiments of this application, the vehicle can send electromagnetic waves to environmental points in the driving environment through environmental sensing devices such as radar, where the radar may include, but is not limited to, millimeter-wave radar and lidar. For example, lidar can be used as an environmental sensing device, and the three-dimensional point cloud in the vehicle's driving environment can be obtained by the echo signals generated when the laser emitted by the lidar illuminates the environmental points. For example, the angular resolution of the lidar can be 0.06°×0.06°, the distance resolution can be 2cm, the scanning FOV can be 120°×25°, and the effective detection distance can be 500m.

[0028] Optionally, in step S101, rasterizing the point cloud space of the point cloud data may include dividing the point cloud space according to the three dimensions of the x, y, and z axes in the coordinate system of the point cloud space to form multiple three-dimensional point cloud grids. Optionally, the size information of the three-dimensional point cloud grids may first be determined using a calibration dataset, wherein the calibration dataset and the point cloud data have the same probability distribution. Then, the point cloud space of the point cloud data is rasterized based on the determined size information to form the multiple three-dimensional point cloud grids. For example, multiple size information of the three-dimensional point cloud grids may be determined using the calibration dataset in an offline stage, and the size information with the best performance may be selected as the determined size information, wherein the best performance may be determined based on one or more of software runtime and hardware utilization.

[0029] In step S103, the point cloud distribution in the point cloud space and multiple three-dimensional point cloud grids are used to selectively enter the sparse operation mode.

[0030] Optionally, in step S103, the point cloud distribution in the point cloud space and multiple three-dimensional point cloud grids of the point cloud data can be input into the sparse operation evaluation model to determine whether to enter the sparse operation mode based on the output of the sparse operation evaluation model. For example, the sparse operation evaluation model can be used to analyze the point cloud distribution in the point cloud space of each frame of point cloud data and the number of point cloud grids generated based on each frame of point cloud data in real time, to compare the software runtime and hardware utilization under the sparse operation mode with other operation modes (e.g., direct three-dimensional convolution operation, sparse convolution operation, etc.) to determine whether to enter the sparse operation mode. When the output of the sparse operation evaluation model determines that the sparse operation mode has been entered, proceed to step S105; when the output of the sparse operation evaluation model determines that the sparse operation mode has not been entered, the point cloud space of the point cloud data can be divided into voxel spaces, and voxel feature maps can be generated by sampling the features of the point cloud data in each voxel space, and three-dimensional convolution or sparse convolution operations can be performed on the voxel feature maps. As can be understood, a voxel is short for volume pixel, which is the smallest unit of digital data segmentation in three-dimensional space, conceptually similar to the smallest unit of pixel in two-dimensional space. For example, the features of point cloud data may include, but are not limited to, the position of the point cloud (such as the x-axis, y-axis, and z-axis coordinates in a three-dimensional coordinate system) and intensity.

[0031] Optionally, the decision to perform a 3D convolution or sparse convolution operation on the voxel feature map can be based on the point cloud density in the voxel space. The point cloud density in the voxel space can be determined by the ratio of the number of points contained in each voxel to the total number of points in the point cloud data. If the point cloud density in the voxel space is greater than or equal to a preset threshold, a 3D convolution operation is performed on the voxel feature map; if the point cloud density is less than the preset threshold, a sparse convolution operation is performed. Since point cloud density represents the number of points contained in a voxel, a higher point cloud density indicates a greater probability that an object exists in that voxel space.

[0032] In step S105, the following operations are performed in sparse operation mode: point cloud gratings containing point cloud data from multiple three-dimensional point cloud gratings are used as target point cloud gratings and a grating mapping relationship is generated based on the corresponding storage address of each target point cloud grating; and a three-dimensional convolution operation is performed on each target point cloud grating based on the grating mapping relationship to generate an output point cloud grating.

[0033] Optionally, in step S105, the size information of the mapping matrix can be determined based on the size information of the point cloud space and the size information of the 3D point cloud raster. A mapping matrix is ​​then created based on the size information of the mapping matrix, and the mapping matrix is ​​used to establish a correspondence between each target point cloud raster and its corresponding storage address to generate a raster mapping relationship. After establishing the correspondence between each target point cloud raster and its corresponding storage address using the mapping matrix, the storage address corresponding to each target point cloud raster in the mapping matrix can be traversed to perform a 3D convolution operation on each target point cloud raster.

[0034] Optionally, the point cloud data processing method according to one or more embodiments of this application may further include determining whether the size of the output point cloud raster is consistent with that of the target point cloud raster, and rearranging the output point cloud raster in response to determining that the size of the output point cloud raster is inconsistent with that of the target point cloud raster, so that the size of the output point cloud raster is consistent with that of the target point cloud raster. Optionally, the point cloud data processing method according to one or more embodiments of this application may further include filtering the output point cloud raster using a mask, wherein the elements in the mask are determined based on the dimension of the feature vector of the point cloud data contained in the target point cloud raster. Optionally, when the dimension of the feature vector of the point cloud data contained in the target point cloud raster is zero, the elements in the mask are set to zero; and when the dimension of the feature vector of the point cloud data contained in the target point cloud raster is not zero, the elements in the mask are set to one.

[0035] Optionally, the point cloud data processing method according to one or more embodiments of this application can be applied to autonomous vehicles. The autonomous vehicle performs real-time target detection on the point cloud data collected by LiDAR to locate the positions of pedestrians, obstacles, and vehicles, and then executes corresponding driving strategies.

[0036] The point cloud data processing method according to one or more embodiments of this application can rasterize the point cloud space of point cloud data according to the optimal point cloud raster size information to form multiple three-dimensional point cloud rasters. Three-dimensional convolution operations are performed on point cloud rasters with point cloud data, while point cloud rasters without point cloud data are skipped and not calculated. Therefore, compared to conventional direct calculation of three-dimensional convolution, the point cloud data processing method according to one or more embodiments of this application does not store invalid point cloud rasters, saving memory, and reduces computational load by not calculating invalid point cloud rasters. Compared to conventional sparse convolution, the point cloud data processing method according to one or more embodiments of this application is simple to implement, eliminating the complex hash table building process, a large number of atomic calculations, and a large number of discontinuous read and write operations. For each target point cloud raster, data reading is continuous and calculation is regular, thereby improving the utilization rate of the tensor computation core on the general image processor and shortening the convolution calculation time.

[0037] The point cloud data processing method according to one or more embodiments of this application, by using point cloud rasters containing point cloud data as target point cloud rasters and performing 3D convolution operations on each target point cloud raster, can transform sparse vector operations into dense matrix operations for point cloud data, thereby saving memory, reducing computational load, shortening computation time, and improving hardware utilization. The point cloud data processing scheme according to one or more embodiments of this application can be applied to various scenarios such as intelligent robot interaction, autonomous driving, and assisted driving, thereby achieving fast and accurate 3D target detection.

[0038] The following will combine Figure 2-3 A point cloud data processing method according to one or more embodiments of this application is further described.

[0039] Figure 2 A schematic diagram illustrating the generation of raster mapping relationships according to one or more embodiments of this application is shown.

[0040] like Figure 2 As shown, the point cloud space of point cloud data can be represented using [D,H,W,C], where D represents the depth of the point cloud space, H represents the height of the point cloud space, w represents the width of the point cloud space, and C represents the dimension of the feature vector of the point cloud data.

[0041] Optionally, after acquiring the point cloud space using lidar, the size information of the 3D point cloud raster can be determined first using a calibration dataset with the same probability distribution as the point cloud data [D]. block H block W block [,C], where D block H represents the depth of the point cloud raster. block W represents the height of the point cloud raster. block Let C represent the width of the point cloud raster, and C represent the dimension of the feature vectors of the point cloud raster. Then, based on the determined 3D point cloud raster size information [D...], ... block H block W block The point cloud space [C,H,W,C] of the point cloud data is rasterized to form [D,C]. ma p,H map W map ] three-dimensional point cloud grids, where [D map H map W map The following formula (1) can be used to calculate it:

[0042]

[0043] Optionally, multiple dimensional information of the 3D point cloud raster can be determined using the calibration dataset during the offline phase, and the dimensional information with the best performance can be selected as the determined dimensional information. The optimal performance can be determined based on one or more factors, such as software runtime and hardware utilization. This allows for independent configuration of the dimensional information of each 3D point cloud raster, improving computational flexibility.

[0044] like Figure 2 As shown, based on the determined size information of the three-dimensional point cloud raster [D block H block W block The point cloud space [D,H,W,C] of the point cloud data is rasterized to form [D,C]. map H map W map After generating a 3D point cloud raster, [D] can be... map H map W map A three-dimensional point cloud raster contains point cloud data (such as...) Figure 2 The point cloud raster (shown as gray dots in the image) is used as the target point cloud raster, and a raster mapping relationship is generated based on the corresponding storage address of each target point cloud raster. For example, a raster of size [D] can be used. map H map W mapThe mapping matrix is ​​used to establish the correspondence between each target point cloud raster and its corresponding storage address to generate the raster mapping relationship. For example, the storage address corresponding to each target point cloud raster can be represented by the address of the first element block[i][0,0,0] of the target point cloud raster, where the address of the first element block[i][0,0,0] of the target point cloud raster can be calculated by the following formula (2):

[0045] block[i][0,0,0]=*(data+i×D block ×H block ×W block ×C) Formula (2)

[0046] Where block[i] represents the i-th target point cloud grid, data represents the starting address of the target point cloud grid, and D block H represents the depth of the target point cloud raster. block W represents the height of the target point cloud raster. block This represents the height of the target point cloud raster, and C represents the dimension of the target point cloud raster's feature vectors. For example... Figure 2 As shown, it can be, for example, through a size [D] map H map W map The mapping matrix is ​​used to establish the correspondence between the target point cloud raster numbered 0, 1...i and the corresponding storage address block[i][0,0,0] to generate the raster mapping relationship.

[0047] pass Figure 2 The process of rasterizing the point cloud space [D,H,W,C] shown can convert unstructured point cloud data into structured three-dimensional point cloud raster, thereby accelerating the convolution process through matrix computation units in subsequent convolution processing, reducing software runtime and improving hardware utilization.

[0048] Figure 3 A schematic diagram of point cloud data processing according to one or more embodiments of this application is shown.

[0049] like Figure 3As shown, after determining the target point cloud raster Bolck 1, Bolck 2... Bolck n and generating the raster mapping relationship, the corresponding storage addresses addr1, addr2, addr3, addr4... of each target point cloud raster in the raster mapping relationship can be traversed to perform a 3D convolution operation on each target point cloud raster. Optionally, an expansion operation can be performed on the target point cloud raster Bolck 1 first. Assuming the storage address of the target point cloud raster Bolck 1 is addr2, performing an expansion operation on the target point cloud raster Bolck 1 can include merging the target point cloud raster stored at the storage address adjacent to the storage address addr2 (e.g., storage address addr3) into the target point cloud raster Bolck 1 to generate the point cloud raster Bolck1'. Next, a 3D convolution operation is performed on the extended point cloud raster Bolck 1' to generate the output point cloud raster Bolck 1'". Understandably, the size of the output point cloud raster Bolck 1' generated after the 3D convolution operation may be smaller, therefore... Figure 3 The point cloud data processing method shown may further include determining whether the size of the output point cloud raster Bolck 1” is consistent with that of the target point cloud raster Bolck 1, and when it is determined that the size of the output point cloud raster Bolck 1” is inconsistent with that of the target point cloud raster Bolck 1, the output point cloud raster Bolck 1” may be rearranged, for example, by combining 8 output point cloud raster Bolck 1” into one point cloud raster, so that the size of the output point cloud raster Bolck 1” is consistent with that of the target point cloud raster Bolck 1.

[0050] like Figure 3 As shown, a mask can also be used to filter the output point cloud raster Bolck 1” to generate a filtered point cloud raster Bolck 1”'. The elements in the mask are determined based on the dimension of the feature vector of the point cloud data contained in the target point cloud raster Bolck 1. Optionally, when the dimension of the feature vector of the point cloud data contained in the target point cloud raster Bolck 1 is zero, the elements in the mask are set to zero; and when the dimension of the feature vector of the point cloud data contained in the target point cloud raster Bolck 1 is not zero, the elements in the mask are set to one.

[0051] Figure 4 A block diagram of a point cloud data processing system according to one or more embodiments of this application is shown.

[0052] like Figure 4As shown, the point cloud data processing system 40 includes a communication unit 410, a memory 420 (e.g., a non-volatile memory such as flash memory, ROM, hard disk, disk, optical disk, etc.), a processor 430, and a computer program 440 stored on the memory 420 and executable on the processor 430.

[0053] The communication unit 410 serves as a communication interface and is configured to establish a communication connection between the point cloud data processing system 40 and external devices or networks (e.g., lidar).

[0054] The memory 420 stores a computer program 440 that can be executed by the processor 430. Furthermore, the memory 420 may also store data generated by the processor 430 when executing the computer program (e.g., target point cloud raster, raster mapping relationships, etc.) and data or commands received from the outside via the communication unit 410.

[0055] Processor 430 is configured to execute computer program 440 to implement point cloud data processing methods according to one or more embodiments of the present application.

[0056] The point cloud data processing system according to one or more embodiments of this application, by using point cloud rasters containing point cloud data as target point cloud rasters and performing 3D convolution operations on each target point cloud raster, can transform sparse vector operations into dense matrix operations for point cloud data, thereby saving memory, reducing computational load, shortening computation time, and improving hardware utilization. The point cloud data processing scheme according to one or more embodiments of this application can be applied to various scenarios such as intelligent robot interaction, autonomous driving, and assisted driving, thereby achieving fast and accurate 3D target detection.

[0057] Additionally, as described above, this application can also be implemented as a computer-readable storage medium including instructions that, when executed, perform a point cloud data processing method according to one aspect of this application.

[0058] Where applicable, the various embodiments provided in this application may be implemented using hardware, software, or a combination of hardware and software. Furthermore, where applicable, without departing from the scope of this application, the various hardware and / or software components described herein may be combined into composite components comprising software, hardware, and / or both. Where applicable, without departing from the scope of this application, the various hardware and / or software components described herein may be divided into sub-components comprising software, hardware, or both. Additionally, where applicable, it is contemplated that software components may be implemented as hardware components, and vice versa.

[0059] The software (such as program code and / or data) according to this application can be stored on one or more computer storage media. It is also contemplated that the software identified herein can be implemented using one or more networked and / or otherwise general-purpose or special-purpose computers and / or computer systems. Where applicable, the order of the various steps described herein can be changed, combined into compound steps, and / or divided into sub-steps to provide the features described herein.

[0060] The embodiments and examples presented herein are provided to best illustrate embodiments of this application and its particular applications, thereby enabling those skilled in the art to implement and use this application. However, those skilled in the art will understand that the above description and examples are provided for ease of illustration and example only. The descriptions presented are not intended to cover all aspects of this application or to limit this application to the precise forms disclosed.

Claims

1. A point cloud data processing method, characterized in that, The method includes the following steps: The point cloud space of the point cloud data is rasterized to form multiple three-dimensional point cloud rasters; Based on the point cloud spatial distribution of the point cloud data and the multiple three-dimensional point cloud grids, selectively enter the sparse operation mode. as well as In the sparse operation mode, the following operations are performed: point cloud rasters containing point cloud data from the plurality of three-dimensional point cloud rasters are used as target point cloud rasters, and a raster mapping relationship is generated based on the corresponding storage address of each target point cloud raster; and a three-dimensional convolution operation is performed on each target point cloud raster based on the raster mapping relationship to generate an output point cloud raster. The point cloud distribution in the point cloud space based on point cloud data and the multiple three-dimensional point cloud grids are used to selectively enter the sparse operation mode, including: The point cloud spatial distribution and the multiple three-dimensional point cloud grids of the point cloud data are input into a sparse operation evaluation model. Based on the output of the sparse operation evaluation model, it is determined whether to enter the sparse operation mode. The method further includes performing the following operations in response to selecting not to enter the sparse operation mode: The point cloud space of the point cloud data is divided into a voxel space; and Voxel feature maps are generated by sampling the features of point cloud data within each voxel space. as well as Perform a 3D convolution operation or a sparse convolution operation on the voxel feature map.

2. The method according to claim 1, wherein rasterizing the point cloud space of the point cloud data to form multiple three-dimensional point cloud grids comprises: The size information of a 3D point cloud raster is determined using a calibration dataset, wherein the calibration dataset has the same probability distribution as the point cloud data; as well as Based on the size information, the point cloud space of the point cloud data is rasterized to form the multiple three-dimensional point cloud grids.

3. The method according to claim 2, wherein generating the raster mapping relationship based on the corresponding storage address of each target point cloud raster includes: The size information of the mapping matrix is ​​determined based on the size information of the point cloud space and the size information of the three-dimensional point cloud raster. Create a mapping matrix based on the size information of the mapping matrix; as well as The mapping matrix is ​​used to establish a correspondence between each target point cloud raster and its corresponding storage address to generate the raster mapping relationship.

4. The method according to claim 1, wherein the method further comprises: Determine whether the size of the output point cloud raster is the same as that of the target point cloud raster; as well as In response to the determination that the output point cloud grid and the target point cloud grid are inconsistent in size, the output point cloud grid is rearranged so that the output point cloud grid and the target point cloud grid are consistent in size.

5. The method according to claim 1, wherein the method further comprises: The output point cloud raster is filtered using a mask, and the elements in the mask are determined based on the dimension of the feature vector of the point cloud data contained in the target point cloud raster.

6. The method of claim 5, wherein the elements in the mask are determined by: When the dimension of the feature vector of the point cloud data contained in the target point cloud raster is zero, the elements in the mask are set to zero; and When the dimension of the feature vector of the point cloud data contained in the target point cloud raster is not zero, the element in the mask is set to one.

7. A point cloud data processing system, characterized in that, The system includes: Memory; A processor coupled to the memory; and A computer program stored in the memory and running on the processor, the execution of which results in the execution of the point cloud data processing method according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed, perform the point cloud data processing method according to any one of claims 1-6.

Images