4D imaging millimeter wave radar point cloud data thickening method and device

Abstract

The invention provides a 4D imaging millimeter wave radar point cloud data thickening method and device, and the method comprises the steps: carrying out the voxelization processing of the point cloud data of a 4D imaging millimeter wave radar, and obtaining a plurality of voxels which at least comprise one point; converting the central point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel; performing semantic segmentation on the RGB image to obtain a foreground image and a background image; determining a plurality of foreground voxels according to the pixels corresponding to the voxels; sampling pixels in the foreground image to obtain a plurality of foreground pixels, and generating a plurality of empty voxels of each foreground pixel in the point cloud coordinate system; processing the foreground pixels and the plurality of corresponding empty voxels by using a processing model to obtain a score of each empty voxel, and taking the empty voxels with the scores exceeding a preset threshold as virtual foreground voxels; and screening the virtual foreground voxels by using the foreground voxels, and taking a screening result as a thickened point cloud data set. According to the invention, the thickened point cloud data contains more target information.

Description

Technical Field

[0001] The present application relates to the technical field of point cloud data processing, and in particular to a method and device for thickening point cloud data of a 4D imaging millimeter wave radar. Background Art

[0002] Compared with traditional automotive radars, 4D imaging millimeter-wave radars contain high-level information and have higher point cloud density. Therefore, multiple datasets providing 4D radar point clouds and 4D radar tensor data have been established. However, the current research on 3D target detection algorithms based on 4D imaging millimeter-wave radar point clouds is still very limited. Most existing methods generally draw on and use modules related to 3D target detection models based on lidar point clouds to perform 3D target detection tasks based on 4D imaging millimeter-wave radar point clouds. However, in general, compared with lidar point clouds, there is a significant gap in the point cloud density and quality of 4D imaging millimeter-wave radars. They are more sparse and contain more noise, which makes the application of modules related to 3D target detection models based on lidar point clouds to 3D target detection tasks based on 4D imaging millimeter-wave radar point clouds often have lower detection accuracy and performance.

[0003] There is currently no effective solution to the sparsity problem of 4D imaging millimeter wave radar point clouds. Summary of the invention

[0004] In view of this, the present application provides a 4D imaging millimeter wave radar point cloud data thickening method and device to solve the above technical problems.

[0005] In a first aspect, an embodiment of the present application provides a 4D imaging millimeter wave radar point cloud data thickening method, comprising:

[0006] Obtain point cloud data and RGB images of 4D imaging millimeter wave radar of the same target scene;

[0007] voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel;

[0008] Performing semantic segmentation on the RGB image to obtain a foreground image and a background image, and determining a plurality of foreground voxels according to the pixels corresponding to each voxel and the foreground image;

[0009] Sampling pixels in the foreground image to obtain a plurality of foreground pixels, and generating a plurality of empty voxels for each foreground pixel in a point cloud coordinate system;

[0010] Processing the foreground pixel and the corresponding multiple empty voxels using the pre-trained processing model to obtain the score of each empty voxel, and taking the empty voxels whose scores exceed a preset threshold as virtual foreground voxels;

[0011] The virtual foreground voxels are screened using the foreground voxels, and the screening results are used as the thickened point cloud dataset.

[0012] Furthermore, the semantic segmentation model adopts the DeepLabv3 network;

[0013] Performing semantic segmentation on the RGB image to obtain a foreground image and a background image includes:

[0014] Use a two-dimensional backbone network to process the RGB image and obtain an RGB feature map;

[0015] The RGB feature map is processed using the pre-trained semantic segmentation model to obtain the foreground image and background image.

[0016] Further, a plurality of foreground voxels are determined according to the pixels corresponding to each voxel and the foreground image, including:

[0017] It is determined whether the pixel corresponding to each voxel belongs to the foreground image. If yes, the voxel is a foreground voxel.

[0018] Further, the processing model includes a first multi-layer perceptron and a second multi-layer perceptron;

[0019] The foreground pixels and the corresponding multiple empty voxels are processed using the pre-trained processing model to obtain the score of each empty voxel, including:

[0020] Using a first multi-layer perceptron to process the coordinates of the center point of each empty voxel, obtaining a first feature of each empty voxel;

[0021] The first feature of each empty voxel is concatenated with the feature of the foreground pixel to obtain the second feature of each empty voxel;

[0022] The second feature of each empty voxel is processed by a second multi-layer perceptron to obtain a score of each empty voxel.

[0023] Furthermore, the virtual foreground voxels are screened using the foreground voxels, and the screening results are used as a thickened point cloud data set; including:

[0024] It is determined whether the virtual foreground voxel is within a sphere of a preset sphere radius with any foreground voxel as the sphere center. If yes, the center point of the virtual foreground voxel is added to the thickened point cloud data set.

[0025] Furthermore, the method further comprises:

[0026] Acquire multiple training sample combinations, wherein the training sample combinations include multiple spatiotemporally matched RGB image samples and point cloud data samples;

[0027] voxelize the point cloud data samples to obtain multiple real voxels containing at least one point;

[0028] Processing the RGB image samples to obtain foreground image samples and background image samples;

[0029] Sampling pixels in the foreground image sample to obtain a plurality of foreground pixel samples, and generating a plurality of empty voxel samples of each foreground pixel sample in the point cloud coordinate system;

[0030] Using a first multi-layer perceptron to process the coordinates of the center point of each empty voxel sample, to obtain a first feature of each empty voxel sample;

[0031] The first feature of each empty voxel sample and the feature of the foreground pixel sample are concatenated to obtain the second feature of each empty voxel sample;

[0032] Using a second multi-layer perceptron to process the second feature of each empty voxel sample, to obtain a prediction score for each empty voxel sample;

[0033] Acquire the real voxel closest to the empty voxel sample, and obtain the probability label of the empty voxel sample by using the Gaussian field of the real voxel and the distance between the empty voxel sample and the real voxel;

[0034] Calculate the loss value using the prediction score and corresponding probability label of each empty voxel sample;

[0035] The loss value is used to update the parameters of the first multilayer perceptron and the second multilayer perceptron.

[0036] In a second aspect, an embodiment of the present application provides a 4D imaging millimeter wave radar point cloud data thickening device, comprising:

[0037] An acquisition unit, used to acquire point cloud data and RGB images of a 4D imaging millimeter wave radar of the same target scene;

[0038] The first processing unit is used to voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel;

[0039] A semantic segmentation unit, configured to perform semantic segmentation on the RGB image to obtain a foreground image and a background image, and determine a plurality of foreground voxels according to pixels corresponding to each voxel and the foreground image;

[0040] A second processing unit is used to sample pixels in the foreground image to obtain a plurality of foreground pixels, and generate a plurality of empty voxels of each foreground pixel in the point cloud coordinate system;

[0041] A third processing unit is used to process the foreground pixel and the corresponding plurality of empty voxels using the pre-trained processing model to obtain a score of each empty voxel, and to use the empty voxels whose scores exceed a preset threshold as virtual foreground voxels;

[0042] The point cloud data thickening unit is used to screen virtual foreground voxels using foreground voxels, and use the screening results as the thickened point cloud data set.

[0043] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the point cloud data thickening method of the embodiment of the present application when executing the computer program.

[0044] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the point cloud data thickening method of the embodiment of the present application is implemented.

[0045] The present application utilizes foreground pixels in a foreground image to thicken point cloud data, so that the thickened point cloud data contains more target information. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 A flowchart of a 4D imaging millimeter wave radar point cloud data thickening method provided in an embodiment of the present application;

[0048] Figure 2 A functional structure diagram of a 4D imaging millimeter wave radar point cloud data thickening device provided in an embodiment of the present application;

[0049] Figure 3 A functional structure diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0052] First, the design concept of the embodiments of the present application is briefly introduced.

[0053] In recent years, with the rise of deep learning and artificial intelligence technologies, intelligent transportation has become an increasingly hot topic in the industry and has received extensive research and attention. 3D object detection plays a key role in intelligent transportation roadside perception systems, and its detection accuracy is directly related to the performance of downstream tasks such as target tracking and prediction. At present, the corresponding 3D object detection methods in roadside perception systems vary due to the different types of sensors used. Common sensors include lidar, cameras, and radars, which have different data structures and characteristics. The data captured by lidar exists in the form of point clouds, which can provide accurate three-dimensional geometric information about the shape, size, and position of objects. The image data collected by the camera is dense and regular, and can provide rich texture and contextual semantic information. However, the high cost of lidar and the fact that cameras are easily disturbed by lighting and weather conditions make them difficult to apply in a wide range of life scenarios in the real world. In contrast, radar is economical and robust to external factors, and has important significance for practical applications.

[0054] In order to improve the 3D target detection accuracy of 4D imaging millimeter-wave radar point clouds, relevant research has explored from the perspective of multimodal fusion, trying to introduce data from other modalities such as camera images, lidar point clouds, etc., and using attention mechanisms and splicing to fuse the original 4D imaging millimeter-wave radar point cloud data for multimodal fusion 3D target detection, but the effect is not good.

[0055] In order to solve the problem that 4D imaging millimeter wave radar point clouds are sparser than lidar point clouds and have poor geometric appearance description capabilities, this application proposes a 4D imaging millimeter wave radar point cloud thickening method, which uses image semantic segmentation and processing models and uses 2D image information to guide the generation of 3D virtual voxels to achieve the thickening of the original 4D imaging millimeter wave radar sparse point cloud.

[0056] This application uses foreground pixels in the foreground image to thicken the point cloud data so that the thickened point cloud data contains more target information. When the thickened point cloud data is applied to downstream detection tasks, the target detection accuracy can be improved.

[0057] After introducing the application scenarios and design concepts of the embodiments of the present application, the technical solutions provided by the embodiments of the present application are described below.

[0058] like Figure 1 As shown, the embodiment of the present application provides a 4D imaging millimeter wave radar point cloud data thickening method, comprising the following steps:

[0059] Step 101: Acquire point cloud data and RGB images of a 4D imaging millimeter wave radar of the same target scene;

[0060] Step 102: voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel;

[0061] Specifically, the point cloud space is divided into regular three-dimensional grids, each three-dimensional grid is a voxel, and the points in the original point cloud are mapped to the three-dimensional grid where they are located.

[0062] For each voxel, its center point is transformed from the world coordinate system to the camera coordinate system for perspective projection to obtain its corresponding pixel coordinates.

[0063] Step 103: semantically segment the RGB image to obtain a foreground image and a background image, and determine a plurality of foreground voxels according to the pixels corresponding to each voxel and the foreground image;

[0064] Specifically, the semantic segmentation model adopts the DeepLabv3 network;

[0065] Performing semantic segmentation on the RGB image to obtain a foreground image and a background image includes:

[0066] Use a two-dimensional backbone network to process the RGB image and obtain an RGB feature map;

[0067] The RGB feature map is processed using the pre-trained semantic segmentation model to obtain the foreground image and background image.

[0068] A plurality of foreground voxels are determined according to the pixels corresponding to each voxel and the foreground image, including:

[0069] It is determined whether the pixel corresponding to each voxel belongs to the foreground image. If yes, the voxel is a foreground voxel.

[0070] Step 104: sampling pixels in the foreground image to obtain a plurality of foreground pixels, and generating a plurality of empty voxels of each foreground pixel in a point cloud coordinate system;

[0071] Step 105: Processing the foreground pixel and the corresponding multiple empty voxels using the pre-trained processing model to obtain the score of each empty voxel, and taking the empty voxels whose scores exceed a preset threshold as virtual foreground voxels;

[0072] Specifically, the processing model includes a first multi-layer perceptron and a second multi-layer perceptron;

[0073] The foreground pixels and the corresponding multiple empty voxels are processed using the pre-trained processing model to obtain the score of each empty voxel, including:

[0074] Using a first multi-layer perceptron to process the coordinates of the center point of each empty voxel, obtaining a first feature of each empty voxel;

[0075] The first feature of each empty voxel is concatenated with the feature of the foreground pixel to obtain the second feature of each empty voxel;

[0076] The second feature of each empty voxel is processed by a second multi-layer perceptron to obtain a score of each empty voxel.

[0077] Step 106: Using the foreground voxels to screen the virtual foreground voxels, and using the screening results as the thickened point cloud data set.

[0078] Wherein, step 106 specifically includes:

[0079] It is determined whether the virtual foreground voxel is within a sphere of a preset sphere radius with any foreground voxel as the sphere center. If yes, the center point of the virtual foreground voxel is added to the thickened point cloud data set.

[0080] Based on the above embodiments, the present application provides a 4D imaging millimeter wave radar point cloud data thickening device, see Figure 2 As shown, the 4D imaging millimeter wave radar point cloud data thickening device 200 provided in the embodiment of the present application at least includes:

[0081] An acquisition unit 201 is used to acquire point cloud data and RGB images of a 4D imaging millimeter wave radar of the same target scene;

[0082] The first processing unit 202 is used to voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel;

[0083] A semantic segmentation unit 203 is used to perform semantic segmentation on the RGB image to obtain a foreground image and a background image, and determine a plurality of foreground voxels according to the pixels corresponding to each voxel and the foreground image;

[0084] The second processing unit 204 is used to sample the pixels in the foreground image to obtain a plurality of foreground pixels, and generate a plurality of empty voxels of each foreground pixel in the point cloud coordinate system;

[0085] The third processing unit 205 is used to process the foreground pixel and the corresponding plurality of empty voxels using the pre-trained processing model to obtain a score of each empty voxel, and to use the empty voxels whose scores exceed a preset threshold as virtual foreground voxels;

[0086] The point cloud data thickening unit 206 is used to screen the virtual foreground voxels using the foreground voxels, and use the screening results as the thickened point cloud data set.

[0087] It should be noted that the principle of solving the technical problem by the 4D imaging millimeter wave radar point cloud data thickening device 200 provided in the embodiment of the present application is similar to the method provided in the embodiment of the present application. Therefore, the implementation of the 4D imaging millimeter wave radar point cloud data thickening device 200 provided in the embodiment of the present application can refer to the implementation of the method provided in the embodiment of the present application, and the repeated parts will not be repeated.

[0088] Based on the above embodiments, the present application also provides an electronic device, referring to Figure 3 As shown, the electronic device 300 provided in the embodiment of the present application includes at least: a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program, the 4D imaging millimeter wave radar point cloud data thickening method provided in the embodiment of the present application is implemented.

[0089] The electronic device 300 provided in the embodiment of the present application may further include a bus 303 connecting different components (including the processor 301 and the memory 302). The bus 303 represents one or more of several types of bus structures, including a memory bus, a peripheral bus, a local bus, and the like.

[0090] The memory 302 may include a readable medium in the form of a volatile memory, such as a random access memory (RAM) 3021 and / or a cache memory 3022 , and may further include a read-only memory (ROM) 3023 .

[0091] The memory 302 may also include a program tool 3025 having a set (at least one) of program modules 3024, the program modules 3024 including but not limited to: an operating subsystem, one or more applications, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0092] The electronic device 300 may also communicate with one or more external devices 304 (e.g., keyboards, remote controls, etc.), and may also communicate with one or more devices that enable a user to interact with the electronic device 300 (e.g., mobile phones, computers, etc.), and / or communicate with any device that enables the electronic device 300 to communicate with one or more other electronic devices 300 (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 305. Furthermore, the electronic device 300 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 306. Figure 3 As shown, the network adapter 306 communicates with other modules of the electronic device 300 via the bus 303. It should be understood that although Figure 3 Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 300, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, disk arrays (Redundant Arrays of Independent Disks, RAID) subsystems, tape drives, and data backup storage subsystems.

[0093] It should be noted that Figure 3 The electronic device 300 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0094] The embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and when the computer instructions are executed by the processor, the 4D imaging millimeter wave radar point cloud data thickening method provided in the embodiment of the present application is implemented. Specifically, the executable program can be built-in or installed in the electronic device 300, so that the electronic device 300 can implement the 4D imaging millimeter wave radar point cloud data thickening method provided in the embodiment of the present application by executing the built-in or installed executable program.

[0095] The 4D imaging millimeter wave radar point cloud data thickening method provided in the embodiment of the present application can also be implemented as a program product, which includes a program code. When the program product can be run on the electronic device 300, the program code is used to enable the electronic device 300 to execute the 4D imaging millimeter wave radar point cloud data thickening method provided in the embodiment of the present application.

[0096] The program product provided in the embodiments of the present application may adopt any combination of one or more readable media, wherein the readable medium may be a readable signal medium or a readable storage medium, and the readable storage medium may be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. Specifically, more specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, RAM, ROM, an erasable programmable read-only memory (EPROM), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0097] The program product provided in the embodiment of the present application may adopt a CD-ROM and include program code, and may also be run on a computing device. However, the program product provided in the embodiment of the present application is not limited thereto. In the embodiment of the present application, the readable storage medium may be any tangible medium containing or storing a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.

[0098] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0099] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application is described in detail with reference to the embodiments, a person skilled in the art should understand that any modification or equivalent replacement of the technical solution of the present application does not depart from the spirit and scope of the technical solution of the present application and should be included in the scope of the claims of the present application.

Claims

1. A 4D imaging millimeter wave radar point cloud data thickening method, characterized in that: include: Obtain point cloud data and RGB images of 4D imaging millimeter wave radar of the same target scene; voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel; Performing semantic segmentation on the RGB image to obtain a foreground image and a background image, and determining a plurality of foreground voxels according to the pixels corresponding to each voxel and the foreground image; Sampling pixels in the foreground image to obtain a plurality of foreground pixels, and generating a plurality of empty voxels for each foreground pixel in a point cloud coordinate system; Processing the foreground pixel and the corresponding multiple empty voxels using the pre-trained processing model to obtain the score of each empty voxel, and taking the empty voxels whose scores exceed a preset threshold as virtual foreground voxels; The virtual foreground voxels are screened using the foreground voxels, and the screening results are used as the thickened point cloud dataset.

2. The method according to claim 1, characterized in that The semantic segmentation model adopts DeepLabv3 network; Performing semantic segmentation on the RGB image to obtain a foreground image and a background image includes: Use a two-dimensional backbone network to process the RGB image and obtain an RGB feature map; The RGB feature map is processed using the pre-trained semantic segmentation model to obtain the foreground image and background image.

3. The method according to claim 1, characterized in that A plurality of foreground voxels are determined according to the pixels corresponding to each voxel and the foreground image, including: It is determined whether the pixel corresponding to each voxel belongs to the foreground image. If yes, the voxel is a foreground voxel.

4. The method according to claim 1, characterized in that: The processing model includes a first multi-layer perceptron and a second multi-layer perceptron; The foreground pixels and the corresponding multiple empty voxels are processed using the pre-trained processing model to obtain the score of each empty voxel, including: Using a first multi-layer perceptron to process the coordinates of the center point of each empty voxel, obtaining a first feature of each empty voxel; The first feature of each empty voxel is concatenated with the feature of the foreground pixel to obtain the second feature of each empty voxel; The second feature of each empty voxel is processed by a second multi-layer perceptron to obtain a score of each empty voxel.

5. The method according to claim 1, characterized in that The virtual foreground voxels are screened using the foreground voxels, and the screening results are used as a thickened point cloud data set; including: It is determined whether the virtual foreground voxel is within a sphere of a preset sphere radius with any foreground voxel as the sphere center. If yes, the center point of the virtual foreground voxel is added to the thickened point cloud data set.

6. The method according to claim 4, characterized in that The method further comprises: Acquire multiple training sample combinations, wherein the training sample combinations include multiple spatiotemporally matched RGB image samples and point cloud data samples; voxelize the point cloud data samples to obtain multiple real voxels containing at least one point; Processing the RGB image samples to obtain foreground image samples and background image samples; Sampling pixels in the foreground image sample to obtain a plurality of foreground pixel samples, and generating a plurality of empty voxel samples of each foreground pixel sample in the point cloud coordinate system; Using a first multi-layer perceptron to process the coordinates of the center point of each empty voxel sample, to obtain a first feature of each empty voxel sample; The first feature of each empty voxel sample and the feature of the foreground pixel sample are concatenated to obtain the second feature of each empty voxel sample; Using a second multi-layer perceptron to process the second feature of each empty voxel sample, to obtain a prediction score for each empty voxel sample; Acquire the real voxel closest to the empty voxel sample, and obtain the probability label of the empty voxel sample by using the Gaussian field of the real voxel and the distance between the empty voxel sample and the real voxel; Calculate the loss value using the prediction score and corresponding probability label of each empty voxel sample; The loss value is used to update the parameters of the first multilayer perceptron and the second multilayer perceptron.

7. A 4D imaging millimeter wave radar point cloud data thickening device, characterized in that: include: An acquisition unit, used to acquire point cloud data and RGB images of a 4D imaging millimeter wave radar of the same target scene; The first processing unit is used to voxelize the point cloud data to obtain a plurality of voxels containing at least one point; convert the center point of each voxel into a camera coordinate system to obtain a pixel corresponding to each voxel; A semantic segmentation unit, configured to perform semantic segmentation on the RGB image to obtain a foreground image and a background image, and determine a plurality of foreground voxels according to pixels corresponding to each voxel and the foreground image; A second processing unit is used to sample pixels in the foreground image to obtain a plurality of foreground pixels, and generate a plurality of empty voxels of each foreground pixel in the point cloud coordinate system; A third processing unit is used to process the foreground pixel and the corresponding plurality of empty voxels using the pre-trained processing model to obtain a score of each empty voxel, and to use the empty voxels whose scores exceed a preset threshold as virtual foreground voxels; The point cloud data thickening unit is used to screen virtual foreground voxels using foreground voxels, and use the screening results as the thickened point cloud data set.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the point cloud data thickening method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the point cloud data thickening method according to any one of claims 1 to 6 is implemented.

Images