Laser radar point cloud dynamic and static segmentation method and device and computer equipment

Abstract

The invention discloses a laser radar point cloud dynamic and static segmentation method and device and computer equipment, and relates to the technical field of point cloud segmentation. The method comprises the following steps: firstly, carrying out voxel segmentation on a target area in a radar scanning range to obtain a voxel map; and then, mapping data points in a plurality of point cloud frames to the voxel map, and calculating to obtain dynamic and static attributes of a target voxel according to the reflectivity of the data points mapped in the target voxel by each point cloud frame. The method can improve the accuracy of point cloud dynamic and static segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud segmentation, and in particular to a laser radar point cloud dynamic and static segmentation method, device and computer equipment. Background Art

[0002] The three-dimensional point cloud data collected by the laser radar when scanning the scene usually includes dynamic points and static points. Point cloud dynamic and static segmentation refers to the identification of dynamic points and static points in the point cloud data. Based on point cloud dynamic and static segmentation, dynamic elements (such as moving vehicles and pedestrians) and static elements (such as buildings and roads) in the scene can be distinguished. It is a preprocessing process for subsequent labeling tasks and perception tasks. For example, in the labeling task of continuous point cloud frames, for the current point cloud frame, you can only focus on the attributes of dynamic points, and the static points directly continue the attributes of the previous frame to improve the labeling efficiency. In perception tasks such as target detection, static points can be ignored and only dynamic points can be processed to increase the computing speed.

[0003] In related technologies, the point cloud dynamic and static segmentation algorithm based on deep learning requires more GPU resources, which is not conducive to the implementation of the algorithm on the vehicle side or other edge computing devices. The method of judging the dynamic and static state of the grid based on whether the grid is occupied, if a dynamic grid is continuously occupied for a long time, it may be misjudged as a static grid, such as in scenes such as busy or congested roads, which will lead to poor accuracy of point cloud dynamic and static segmentation. Summary of the invention

[0004] In view of this, the present application provides a laser radar point cloud dynamic and static segmentation method, device and computer equipment, the main purpose of which is to solve the technical problem of low accuracy of point cloud dynamic and static segmentation in related technologies.

[0005] According to a first aspect of the present invention, a laser radar point cloud dynamic and static segmentation method is provided, the method comprising:

[0006] Perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map;

[0007] The data points in a plurality of point cloud frames are mapped to the voxel map, and the dynamic and static properties of the target voxel are calculated according to the reflectivity of the data points mapped in the target voxel by each point cloud frame.

[0008] In some embodiments, before mapping the data points in multiple point cloud frames to the voxel map and calculating the dynamic and static properties of the target voxel based on the reflectivity of the data points mapped in each point cloud frame in the target voxel, the method also includes: mapping a preset first frame number of point cloud frames to the voxel map to obtain a mapping relationship between the data points in each point cloud frame and each voxel; for any voxel, according to the mapping relationship, counting the frame number of the point cloud frames in which the number of data points mapped in any voxel is greater than zero, recorded as a second frame number; calculating the ratio of the second frame number to the first frame number, and if the ratio of the second frame number to the first frame number exceeds a preset first threshold, marking the voxel as a target voxel.

[0009] In some embodiments, the dynamic and static properties of the target voxel are calculated based on the reflectivity of the data points mapped in the target voxel of each point cloud frame, including: for each point cloud frame, calculating the reflectivity of the data points mapped in the target voxel by the point cloud frame to obtain a plurality of reflectivities corresponding to the target voxel; calculating a discrete eigenvalue of the target voxel based on the plurality of reflectivities corresponding to the target voxel; and determining the dynamic and static properties of the target voxel based on the discrete eigenvalue of the target voxel.

[0010] In some embodiments, determining the dynamic and static properties of the target voxel based on the discrete eigenvalue of the target voxel includes: when the discrete eigenvalue of the target voxel is less than a preset second threshold, determining that the target voxel is a static voxel, otherwise, determining that the target voxel is a dynamic voxel.

[0011] In some embodiments, the step of calculating the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame includes:

[0012] Calculate the discrete eigenvalue of the target voxel and the discrete eigenvalue of each bounding box voxel corresponding to the target voxel, wherein the discrete eigenvalue is calculated based on the reflectivity of the data point mapped in the voxel of each point cloud frame; calculate the weight of the target voxel and multiple bounding box voxels based on the distance between each bounding box voxel and the target voxel; obtain the weighted discrete eigenvalue of the target voxel based on the discrete eigenvalue of the target voxel and the weight of the target voxel, as well as the discrete eigenvalues ​​of multiple bounding box voxels and the weights of multiple bounding box voxels; when the weighted discrete eigenvalue is less than a preset second threshold, determine that the target voxel is a static voxel, otherwise, determine that the target voxel is a dynamic voxel.

[0013] In some embodiments, after calculating the dynamic and static properties of the target voxel, the method further includes: mapping the data points in the point cloud frame to be segmented to the voxel map, and determining the dynamic and static properties of the data points in the point cloud frame to be segmented based on the dynamic and static properties of the voxels in the voxel map.

[0014] In some embodiments, the method further includes: reacquiring a plurality of the point cloud frames at a preset time interval, recalculating the dynamic and static properties of the target voxels based on the reacquired plurality of the point cloud frames, and updating the dynamic and static properties of the target voxels in the voxel map.

[0015] According to a second aspect of the present invention, a laser radar point cloud dynamic and static segmentation device is provided, the device comprising:

[0016] A voxel map building module is used to perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map;

[0017] The voxel static and dynamic attribute determination module is used to map the data points in multiple point cloud frames to the voxel map, and calculate the static and dynamic attributes of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame.

[0018] According to a third aspect of the present invention, there is provided a storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the above-mentioned laser radar point cloud dynamic and static segmentation method.

[0019] According to a fourth aspect of the present invention, there is provided a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned laser radar point cloud dynamic and static segmentation method when executing the program.

[0020] The present invention provides a laser radar point cloud dynamic and static segmentation method, device and computer equipment, first voxel segmentation of the target area within the radar scanning range to obtain a voxel map; then the data points in multiple point cloud frames are mapped to the voxel map, and the dynamic and static properties of the target voxels are calculated according to the reflectivity of the data points mapped in the target voxels of each point cloud frame. The laser radar point cloud segmentation method provided by the present application voxelizes the target area to establish a voxel map, first identifies the dynamic and static properties of the target voxels in the voxel map, and then determines the dynamic and static properties of the data points in the point cloud frame according to the dynamic and static properties of the target voxels. Among them, when identifying the dynamic and static properties of the target voxels, the reflectivity of the data points mapped in the voxels of each point cloud frame is used to analyze whether the object corresponding to each point cloud frame in the voxel is the same object, so as to determine the dynamic and static properties of the target voxels. The probability of dynamic voxels being misjudged as static voxels due to long-term occupancy is reduced, and the accuracy of point cloud dynamic and static segmentation can be improved.

[0021] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 A schematic flow chart of a laser radar point cloud dynamic and static segmentation method provided by an embodiment of the present invention is shown;

[0024] Figure 2 A schematic structural diagram of a laser radar point cloud dynamic and static segmentation device provided by an embodiment of the present invention is shown;

[0025] Figure 3 A structural schematic diagram of a computer device for implementing a laser radar point cloud dynamic and static segmentation method provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. In the various embodiments shown below, the first, second and various digital numbers are only used to distinguish for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0027] The three-dimensional point cloud data collected by the laser radar scanning the scene usually includes dynamic points and static points. The dynamic points usually correspond to dynamic elements in the scene, such as moving vehicles and pedestrians, and the static points usually correspond to static elements in the scene, such as buildings and roads. Point cloud dynamic and static segmentation refers to the identification of dynamic points and static points in point cloud data, which is a preprocessing process for subsequent labeling and perception tasks. For example, in the labeling task of continuous point cloud frames, by using point cloud dynamic and static segmentation, for the current point cloud frame, only the attributes of dynamic points can be focused on, and the static points directly continue the attributes of the previous frame to improve the labeling efficiency. In perception tasks such as target detection, by using point cloud dynamic and static segmentation, static points can be ignored and only dynamic points can be processed to improve the computing speed.

[0028] In related technologies, the point cloud static and dynamic segmentation method based on deep learning requires more GPU resources, which is not conducive to the implementation of the algorithm on the vehicle side or other edge computing devices. The method of judging the static and dynamic state of the grid based on whether the grid is occupied is not very accurate in dynamic environments. For example, on a busy or congested road, as different vehicles continue to pass through, a grid may continue to be occupied for a long time. At this time, the grid will be judged as a static grid. The accuracy of point cloud static and dynamic segmentation using this method is poor.

[0029] Based on the above problems, the present application embodiment provides a laser radar point cloud dynamic and static segmentation method. In one embodiment, Figure 1 As shown, the method comprises the following steps:

[0030] 101. Perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map.

[0031] In this embodiment, the radar scanning area is determined based on the point cloud data obtained by radar scanning, and the radar scanning area is a three-dimensional space area. Based on the radar scanning area, the target area is determined, and the target area is a part of the radar scanning area. The target area can also be called ROI (region of interest). After determining the target area, dynamic and static segmentation can be performed only on the point cloud of the target area to reduce the amount of calculation. Of course, the target area can also be the entire area within the radar scanning range, and the target area can be set according to actual conditions and needs.

[0032] After determining the target area, the target area is voxelized. Voxels can be understood as similar to pixels. Pixels are the basic units in two-dimensional space, and voxels refer to the basic units in three-dimensional space. Voxels can be represented by three-dimensional coordinates. Voxelizing the target area and establishing a voxel map means dividing the target area into multiple voxels and assigning coordinates to each voxel. For example, determine the scale of the voxel (or resolution), divide the target area into multiple three-dimensional grids according to the scale of the voxel, each three-dimensional grid corresponds to a voxel, and then establish a coordinate system based on the target area, determine the coordinates of each voxel, and obtain a voxel map. Among them, the coordinates of the voxel can be represented by the coordinates of the center position of the voxel (x, y, z).

[0033] Generally, the smaller the voxel scale, the more accurate the calculation result, but the corresponding amount of calculation will be greater; the larger the voxel scale, the smaller the amount of calculation, but the accuracy of the calculation result will be lower. The size of the voxel scale can be determined according to actual computing resources and needs. In the embodiment of the present application, a smaller voxel scale is selected to ensure that only one type of object point cloud is included in a voxel at the same time with a higher probability.

[0034] In this embodiment, by voxelizing the target area, the dynamic and static properties of the voxels in the voxel map are first determined, and then the data points in the point cloud frame to be segmented are mapped to the voxel map. In subsequent calculations, the dynamic and static properties of the data points in the point cloud frame to be segmented can be determined based on the dynamic and static properties of the voxels. In this way, the dynamic and static properties of the point cloud are identified without the need for a deep learning algorithm, and the amount of computation is low, which does not occupy too many computing resources and is easy to implement on the vehicle side or other edge computing devices.

[0035] 102. Map data points in a plurality of point cloud frames to the voxel map, and calculate the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel by each point cloud frame.

[0036] In this embodiment, after the voxel map is established, the dynamic and static properties of the voxels in the voxel map are determined through multiple point cloud frames. Specifically, based on the point cloud data collected by the radar over a period of time, multiple continuous point cloud frames corresponding to the target area are extracted, wherein the data points in the point cloud frames carry reflectivity information. Each point cloud frame is mapped to the voxel map respectively to obtain a mapping relationship between the data points in each point cloud frame and the target voxel in the voxel map. The target voxel can be all the voxels corresponding to the target area, or it can be at least one user-defined voxel corresponding to a special area.

[0037] According to the mapping relationship, we can determine which voxel in the voxel map any data point in the point cloud frame falls into, and we can also determine which point cloud frames have data points in any voxel, which point cloud frames do not have data points in this voxel, and which specific data points fall into this voxel. Then, we can calculate the dynamic and static properties of the voxel based on the reflectivity of the data points mapped in each point cloud frame in the voxel.

[0038] The reflectivity of the data point in the point cloud frame refers to the percentage of the radiation energy reflected by the surface of the object to the total radiation energy, which reflects the reflectivity of the surface of the object to the incident laser. Generally speaking, different objects have different reflectivities to the incident laser, and different positions of the same object have different reflectivities to the incident laser. Based on this characteristic, by analyzing the reflectivity of the data points falling into a certain voxel, it can be determined whether the object corresponding to the data points in different point cloud frames in the voxel is the same object. If it is the same object, it can be further determined that the dynamic and static attributes of the voxel are static attributes. If it is not the same object, it is determined that the dynamic and static attributes of the voxel are dynamic attributes. In this way, on a busy or congested road, as different vehicles continue to pass through, a grid may continue to show an occupied state for a long time, but due to the different surface materials of different vehicles, the reflectivity of the data points mapped to the voxel fluctuates greatly. It can be considered that the grid is in a state of being continuously passed through by different objects, and the voxel can be defined as a dynamic voxel, avoiding the dynamic and static segmentation error caused only by the occupancy state in the traditional method.

[0039] Among them, if the reflectivity distribution of data points falling into a certain voxel in different point cloud frames tends to be consistent, it can be determined that the objects corresponding to these data points are the same object. If the reflectivity distribution is relatively messy, it can be determined that the objects corresponding to these data points are not the same object. Whether the reflectivity tends to be consistent can be determined by statistical analysis, such as calculating the variance of the reflectivity of the data points in the voxel and determining whether the reflectivity tends to be consistent based on the degree of discreteness; for example, calculating the entropy value of the reflectivity and determining whether the reflectivity tends to be consistent based on the degree of disorder of the reflectivity, etc.

[0040] For example, 100 frames of point cloud data are mapped to a target voxel, and the reflectivity of the target voxel in each frame is taken as the average reflectivity of all data points mapped to the target voxel in the frame. If the discrete entropy value of the reflectivity of the target voxel in each frame within 100 frames is greater than a certain threshold, the target voxel can be defined as a dynamic voxel.

[0041] In this embodiment, the corresponding voxels can be selected as target voxels in the voxel map according to the point cloud data points for dynamic and static segmentation as needed, and the number of target voxels can be one or more. Then, according to the reflectivity of the data points mapped in the target voxels of each point cloud frame, the distribution of the reflectivity of the target voxels relative to different point cloud frames is calculated. According to the distribution of the reflectivity, it can be analyzed whether the objects corresponding to each point cloud frame in the target voxel are the same object, and then it can be determined whether the target voxel is a dynamic voxel or a static voxel. After determining the dynamic and static attributes of the target voxel, the dynamic and static attributes of the target voxel can be assigned to the point cloud data points corresponding to the target voxels according to the mapping relationship between the point cloud data points to be segmented and the target voxels.

[0042] In the above embodiment, the method of calculating the dynamic and static properties of voxels by mapping the reflectivity of data points in voxels through multiple point cloud frames can improve the accuracy of point cloud dynamic and static segmentation and reduce the probability of misjudgment in the case where data points continue to fall into voxels due to the continuous movement of different objects in a dynamic environment.

[0043] The laser radar point cloud dynamic and static segmentation method provided in this embodiment first performs voxel segmentation on the target area within the radar scanning range to obtain a voxel map; then maps the data points in multiple point cloud frames to the voxel map, and calculates the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame. The laser radar point cloud segmentation method provided in this application performs voxelization on the target area to establish a voxel map, first identifies the dynamic and static properties of the target voxel in the voxel map, and then determines the dynamic and static properties of the data points in the point cloud frame according to the dynamic and static properties of the target voxel. Among them, when identifying the dynamic and static properties of the target voxel, the reflectivity of the data points mapped in the voxel of each point cloud frame is used to analyze whether the object corresponding to each point cloud frame in the voxel is the same object, so as to determine the dynamic and static properties of the target voxel. The probability of dynamic voxels being misjudged as static voxels due to long-term occupancy is reduced, and the accuracy of point cloud dynamic and static segmentation can be improved.

[0044] Furthermore, in order to fully illustrate the implementation process of this embodiment, the specific implementation methods of the above-mentioned embodiments are refined and expanded below. Specifically, in one embodiment, before step 102 maps the data points in multiple point cloud frames to the voxel map, and calculates the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in each point cloud frame in the target voxel, the method also includes: mapping a preset first number of point cloud frames to the voxel map to obtain a mapping relationship between the data points in each point cloud frame and each voxel; for any voxel, according to the mapping relationship, counting the number of point cloud frames in which the number of data points mapped in any voxel is greater than zero, recorded as the second number of frames; calculating the ratio of the second number of frames to the first number of frames, if the ratio of the second number of frames to the first number of frames exceeds a preset first threshold, the voxel is marked as a target voxel.

[0045] In the above embodiment, before the target voxel is segmented into static and dynamic states, the target voxel range is further narrowed. That is, the occupancy state judgment method is first used to filter out voxels that have not been occupied for a long time from the static voxels, and the remaining voxels are defined as target voxels, and then the target voxel is segmented into static and dynamic states, that is, step S102.

[0046] Specifically, the static and dynamic properties of the voxel can be preliminarily determined by the number of data points mapped in each point cloud frame in the voxel. For the voxels that are preliminarily determined to be static properties, the static and dynamic properties of the voxel are further determined based on the reflectivity. For example, for a certain voxel, if only a small number (for example, no more than one tenth) of the point cloud frames in the corresponding multiple point cloud frames have data points falling into the voxel, then the probability that the voxel is a dynamic voxel is relatively high, and the static and dynamic properties of the voxel can be determined as a dynamic voxel; if more than a preset number of point cloud frames in the corresponding multiple point cloud frames have data points falling into the voxel, then the static and dynamic properties of the voxel may be a static voxel or a dynamic voxel, and the dynamic and static properties of the voxel need to be further determined based on the reflectivity of the data points in each point cloud frame that fall into the voxel.

[0047] Based on this, before calculating the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel by each point cloud frame in step 102, it is also possible to first select in the voxel map whether the target voxel whose dynamic and static properties need to be further determined by reflectivity according to the amount of data mapped in each voxel by each point cloud frame.

[0048] Specifically, after establishing the voxel map, the preset first number of point cloud frames corresponding to the target area can be obtained, and each point cloud frame can be mapped to the voxel map to obtain the mapping relationship between the data points in each point cloud frame and each voxel in the voxel map. Among them, the first number of frames can be set according to actual conditions, for example, it can be set to 100 frames, 200 frames, etc. The number of point cloud frames used to determine the dynamic and static properties of voxels can also be set by a preset acquisition time, for example, collecting point cloud frames within 1 second as the first number of point cloud frames.

[0049] After obtaining the mapping relationship between the data points of each point cloud frame and each voxel, according to the mapping relationship, the point cloud frames whose number of data points mapped in the target voxels is not zero are screened out from the point cloud frames of the first frame number, and recorded as target point cloud frames. In other words, if a point cloud frame has a data point falling into a voxel, then the point cloud frame is the target point cloud frame corresponding to the voxel. At the same time, the number of target point cloud frames is counted to obtain the second frame number. For example, for any voxel (x, y, z), the second frame number corresponding to the voxel can be recorded as map (x, y, z), and each time a target point cloud frame is identified, the value of map (x, y, z) is increased by 1 until all the point cloud frames of the first frame number are traversed.

[0050] By comparing the proportion of the point cloud frame of the second frame number relative to the point cloud frame of the first frame number, the dynamic and static properties of the voxel are preliminarily determined. For example, the ratio of the second point cloud frame to the first point cloud frame can be calculated, and the first threshold can be preset. If the ratio of the second point cloud frame to the first point cloud frame is greater than the first threshold, the dynamic and static properties of the voxel are preliminarily determined to be a potential static voxel, otherwise the voxel is determined to be a dynamic voxel. Alternatively, the product of the first frame number and the first threshold is calculated. If the second frame number exceeds the product of the first frame number and the first threshold, the voxel is preliminarily determined to be a potential static voxel, otherwise the voxel is determined to be a dynamic voxel. Usually, the first frame number and the first threshold are basically fixed values ​​after being set, or at least fixed values ​​for a period of time. Therefore, usually at least for a period of time, the product of the first frame number and the first threshold only needs to be calculated once. Therefore, when the method of comparing the product of the first frame number and the first threshold with the second frame number is used to preliminarily determine the dynamic and static properties of each voxel, only a comparison operation needs to be performed, and this method has a smaller amount of calculation. In a specific example, the first frame number can be recorded as total_frame, and the second frame number can be recorded as map(x,y,z). If map(x,y,z)>total_frame*first threshold, the voxel is marked as a potential static voxel, otherwise the voxel is marked as a dynamic voxel. The first threshold can be set according to actual conditions, for example, it can be set to 25%, 30%, etc.

[0051] In the above embodiment, potential static voxels that may be static voxels or dynamic voxels are first screened out according to the number of data points mapped in the voxels of each point cloud frame, and then the potential static voxels are marked as target voxels, that is, only the dynamic and static properties of the potential static voxels are further determined by reflectivity. In this way, the amount of calculation can be reduced, the response speed can be improved, and the probability of misjudging dynamic voxels as static voxels can be reduced, thereby improving the accuracy of point cloud dynamic and static segmentation.

[0052] In one embodiment, after determining the target voxel, the dynamic and static properties of the target voxel are calculated based on the reflectivity of the data points mapped in the target voxel by each point cloud frame. This can be specifically achieved in the following way: for each point cloud frame, the reflectivity of the data points mapped in the target voxel by the point cloud frame is calculated to obtain multiple reflectivities corresponding to the target voxel; based on the multiple reflectivities corresponding to the target voxel, the discrete eigenvalues ​​of the target voxel are calculated; and the dynamic and static properties of the target voxel are determined based on the discrete eigenvalues ​​of the target voxel.

[0053] In the above embodiment, for each point cloud frame, the reflectivity of the target voxel in the frame can be calculated based on the reflectivity carried by each data point mapped in the target voxel by the point cloud frame. After filtering out the case where the reflectivity corresponding to the target voxel is 0, the discrete eigenvalues ​​of the data points in the target voxel are calculated based on the reflectivity corresponding to the target voxel in multiple frames. It is also possible to calculate the reflectivity corresponding to each target point cloud frame in multiple frames only for the target point cloud frame. At this time, because the target point cloud frame is a point cloud frame in which the number of data points mapped in the target voxel is not zero, there is no need to filter, and the discrete eigenvalues ​​can be directly calculated.

[0054] The discrete eigenvalue of the target voxel is used to characterize the discrete degree of reflectivity of the data points in the target voxel corresponding to the multiple point cloud frames, that is, the discrete degree of the object types in the target voxel corresponding to the multiple point cloud frames. The discrete eigenvalue can be calculated by statistical methods such as variance and entropy. In this embodiment, it is assumed that there are n point cloud frames in total, and the reflectivity of the target voxel in the first frame is i1, the reflectivity in the second frame is i2, and so on, the reflectivity in the nth frame is i n . Then i1 to i n is a plurality of reflectivities corresponding to the target voxel. Among them, the reflectivity of the target voxel in the nth frame can be taken as the average reflectivity of all data points mapped to the target voxel by the nth frame point cloud. Taking the discrete eigenvalue as Shannon entropy as an example, the consistency of the reflectivity of the data points falling into the target voxel is analyzed. Specifically, after the reflectivity corresponding to the target voxel is calculated for each point cloud frame, the multiple reflectivities corresponding to the target voxel in multiple frames are obtained, and the probability distribution of the reflectivity corresponding to the target voxel is further calculated based on the multiple reflectivities. Then, the Shannon entropy of the target voxel is calculated based on the probability distribution of the reflectivity of the target voxel. Of course, the reflectivity of the target voxel in the nth frame taking the average reflectivity of all data points in the voxel is only one implementation method. In actual situations, the reflectivity of the target voxel in the nth frame can also take the reflectivity or average reflectivity of the data points at a specific position in the center of the voxel mapped to the target voxel by the nth frame point cloud, that is, the influence of the reflectivity of the data points at the edge of the voxel is eliminated.

[0055] Among them, the calculation formula of the Shannon entropy of the target voxel is as follows:

[0056] H(x0,y0,z0)=∑Plog(P);

[0057] In the above formula, H(x0, y0, z0) is the Shannon entropy of the target voxel (x0, y0, z0), and P is the probability distribution of the reflectivity of the target voxel in multiple frames.

[0058] In the above calculation, because the discrete eigenvalues ​​contain entropy information of the reflectivity, and the entropy value can reflect the degree of chaos, the degree of chaos of the object types corresponding to different point cloud frames in the target voxels can be determined based on the discrete eigenvalues, and then the dynamic and static properties of the target voxels can be determined.

[0059] In one embodiment, the dynamic and static properties of the target voxel are determined based on the discrete eigenvalue of the target voxel, which can be specifically achieved in the following manner: when the discrete eigenvalue of the target voxel is less than a preset second threshold, the target voxel is determined to be a static voxel, otherwise, the target voxel is determined to be a dynamic voxel.

[0060] In the above embodiment, the Shannon entropy of the target voxel can be directly used as a discrete eigenvalue, and a second threshold value can be set. When the discrete eigenvalue is less than the second threshold value, it indicates that the reflectivity values ​​of the data points of different point cloud frames in the target voxel tend to be consistent. At this time, the dynamic and static attributes of the target voxel can be determined as a static voxel; if the discrete eigenvalue is greater than the second threshold value, it indicates that the reflectivity values ​​of the data points of different point cloud frames in the target voxel are relatively chaotic and disordered. At this time, the dynamic and static attributes of the target voxel can be determined as a dynamic voxel, wherein the second threshold value can be set according to actual conditions. For example, the second threshold value can be 0.2, 0.3, etc.

[0061] In one embodiment, the static and dynamic properties of the target voxel are calculated according to the reflectivity of the data points mapped in the target voxel of each point cloud frame, which can be specifically implemented in the following manner: the discrete eigenvalue of the target voxel and the discrete eigenvalue of each bounding box voxel corresponding to the target voxel are calculated, wherein the discrete eigenvalue is calculated according to the reflectivity of the data points mapped in the voxel of each point cloud frame; the weight of the target voxel and multiple bounding box voxels are calculated according to the distance between each bounding box voxel and the target voxel; the weighted discrete eigenvalue of the target voxel is obtained according to the discrete eigenvalue of the target voxel and the weight of the target voxel, as well as the discrete eigenvalues ​​of multiple bounding box voxels and the weights of multiple bounding box voxels; when the weighted discrete eigenvalue is less than a preset second threshold, the target voxel is determined to be a static voxel, otherwise, the target voxel is determined to be a dynamic voxel.

[0062] In the above embodiment, it is considered that the dynamic and static attributes of each data point are related to the dynamic and static attributes of the surrounding data points. For example, among the 27 voxels of 3×3×3, when the voxels around the central voxel (bounding box voxels) are all dynamic voxels, there is a high probability that the central voxel is also a dynamic voxel. Therefore, in order to further improve the accuracy of the calculation of the dynamic and static attributes of the target voxel, when calculating the dynamic and static attributes of the target voxel, the dynamic and static attributes of multiple bounding box voxels on the bounding box of the target voxel can be introduced.

[0063] Specifically, based on a method similar to the above-mentioned embodiment, the discrete eigenvalues ​​of the target voxel and the discrete eigenvalues ​​of each bounding box voxel corresponding to the target voxel are calculated respectively. The bounding box voxel can be a voxel in at least one direction of up and down, left and right, front and back, and circumferential directions adjacent to the target voxel. For example, with the target voxel as the center, 3×3×3 bounding box voxels are delineated, and the central voxel is surrounded by a total of 26 bounding box voxels. Then, according to the distance between the bounding box voxel and the target voxel, the correlation between each bounding box voxel and the target voxel is calculated by a Gaussian function, and a weight is assigned to each bounding box voxel. According to the weight distribution of the target voxel and each bounding box voxel, the discrete eigenvalues ​​of the target voxel and each bounding box voxel are weighted and summed to obtain the weighted discrete eigenvalue of the target voxel. The dynamic and static properties of the target voxel are determined according to the weighted discrete eigenvalue of the target voxel.

[0064] Among them, the bounding box voxel of the target voxel is the voxel adjacent to the target voxel in the voxel map. For example, the number of bounding box voxels of the target voxel in the three-dimensional space is 3×3×3-1=26. The calculation method of the discrete eigenvalue of the bounding box voxel is the same as the calculation method of the discrete eigenvalue of the target voxel, and the above embodiment can be referred to. The correlation between the bounding box voxel and the target voxel is calculated by the Gaussian function, and the calculated correlation is normalized to obtain the weight of the target voxel and each bounding box voxel. The calculation formula is as follows:

[0065]

[0066]

[0067] In the above two formulas, n is an integer between [0, 26], and w is n (x n ,y n ,z n ) is the coordinate (x n ,y n ,z n ) is related to the target voxel. When n is 0, the voxel (x n ,y n ,z n ) is the target voxel. When n is any value from 1 to 26, the voxel (x n ,y n ,z n ) is the bounding box voxel. When n=0, the correlation is the largest, which means that the target voxel is most correlated with itself, and the correlation with the surrounding bounding box voxels depends on the distance. σ is an adjustable parameter, which can be set according to the actual situation, for example, it can be set to 1. q n (x n ,y n,z n ) is the coordinate (x n ,y n ,z n ) is the normalized weight of the voxels.

[0068] Then, the discrete eigenvalues ​​of the target voxel and the bounding box voxel are weighted and summed according to the weights to obtain the weighted discrete eigenvalue of the target voxel. Taking the Shannon entropy characterization of discrete eigenvalues ​​as an example, the weighted discrete eigenvalue of the target voxel is the weighted Shannon entropy, and the calculation formula is as follows:

[0069]

[0070] in, is the weighted Shannon entropy of the target voxel (x0, y0, z0), H n (x n ,y n ,z n ) is the coordinate (x n ,y n ,z n ) of the voxels, q n (x n ,y n ,z n ) is the coordinate (x n ,y n ,z n ) is the normalized weight of the voxels.

[0071] After the weighted discrete eigenvalue of the target voxel is calculated, similarly, when the weighted discrete eigenvalue is less than the preset second threshold, the dynamic and static attribute of the target voxel is determined to be a static voxel, and when the weighted discrete eigenvalue is greater than the preset second threshold, the dynamic and static attribute of the target voxel is determined to be a dynamic attribute. In the above embodiment, the second threshold can be set according to actual conditions, for example, the second threshold can be 0.2, 0.3, etc.

[0072] In the above embodiment, the weighted discrete eigenvalue of the target voxel is calculated based on the discrete eigenvalue of the target voxel and the discrete eigenvalue of the bounding box voxel. The method of determining the dynamic and static properties of the target voxel by the weighted discrete eigenvalue takes into account the dynamic and static properties of the voxels around the target voxel, that is, the dynamic and static properties of the surrounding data points. Therefore, the determination of the dynamic and static properties of the target voxel will be more accurate and can improve robustness.

[0073] In one embodiment, after calculating the dynamic and static properties of the target voxel, the method further includes: mapping the data points in the point cloud frame to be segmented to the voxel map, and determining the dynamic and static properties of the data points in the point cloud frame to be segmented based on the dynamic and static properties of the voxels in the voxel map.

[0074] In the above embodiment, after determining the static and dynamic attributes of the voxels in the voxel map, the point cloud frame to be segmented is mapped to the voxel map, and according to the mapping relationship between the data points in the point cloud frame to be segmented and the voxels in the voxel map, the static and dynamic attributes of the voxels are determined as the static and dynamic attributes of the data points corresponding to the voxels. For example, if a data point in the point cloud frame to be segmented falls into a dynamic voxel in the voxel map, the static and dynamic attributes of the data point are dynamic points, and if it falls into a static voxel, the static and dynamic attributes of the data point are static points.

[0075] Specifically, for the data points in the point cloud frame to be segmented, according to the mapping relationship between the data points in the point cloud frame to be segmented and the voxels in the voxel map, the dynamic and static attributes of the voxels corresponding to the data points are determined as the dynamic and static attributes of the data points; each data point in the point cloud frame to be segmented is traversed to obtain the dynamic and static attributes of each data point in the point cloud frame to be segmented; or, all the dynamic voxels and all the static voxels in the voxel map are obtained, and according to the mapping relationship between the data points in the point cloud frame to be segmented and the voxels in the voxel map, the data points corresponding to all the dynamic voxels are marked as dynamic points, and the data points corresponding to all the static voxels are marked as static points, so as to obtain the dynamic and static attributes of each data point in the point cloud frame to be segmented.

[0076] In the above embodiment, when the point cloud frame to be segmented is subjected to static and dynamic segmentation, the point cloud frame to be segmented can be mapped to the voxel map, and the static and dynamic attributes of the data points can be determined according to the mapping relationship between the data points in the point cloud frame to be segmented and the voxels in the voxel map and the static and dynamic attributes of each voxel. Specifically, for any data point in the point cloud frame to be segmented, the voxel corresponding to the data point can be determined, and the static and dynamic attributes of the voxel corresponding to the data point can be determined as the static and dynamic attributes of the data point, and each data point in the point cloud frame to be segmented can be traversed to obtain the static and dynamic attributes of each data point. It is also possible to first determine all static voxels (or dynamic voxels) in the voxel map, determine the data points corresponding to all static voxels in the point cloud frame to be segmented as static points (or dynamic points), and determine the remaining data points in the point cloud frame to be segmented as dynamic points (or static points), so as to obtain the static and dynamic attributes of each data point in the point cloud frame to be segmented. In this way, the static and dynamic attributes of data points can be determined in batches, and the response speed is faster.

[0077] In one embodiment, the laser radar point cloud dynamic and static segmentation method provided in the present application also includes: re-acquiring multiple point cloud frames at a preset time interval, recalculating the dynamic and static properties of the target voxels based on the re-acquired multiple point cloud frames, and updating the dynamic and static properties of the target voxels in the voxel map.

[0078] In the above embodiment, the static and dynamic attributes of each voxel in the voxel map usually do not change much within a period of time. Therefore, a time interval can be preset. Within the time interval, the static and dynamic attributes of the voxel only need to be calculated once. After the static and dynamic attributes of the voxel are calculated, all point cloud frames within this period are segmented based on the static and dynamic attributes of the voxel in the current time period. For example, the preset time interval can be set to 1 minute, and the first 100 point cloud frames collected within 1 minute (or multiple point cloud frames collected in the first second) are selected to calculate the static and dynamic attributes of each voxel, and then the static and dynamic attributes of each voxel calculated this time are segmented for all point cloud frames collected within this minute. Considering that the static and dynamic attributes of the voxel may also change, the static and dynamic attributes of each voxel can be recalculated at every preset time interval, and then the static and dynamic attributes of the point cloud can be segmented based on the recalculated static and dynamic attributes of the voxel. In the above example, 100 frames of point cloud frames are collected every 1 minute to calculate the dynamic and static attributes of each voxel, and the dynamic and static attributes of each voxel in the voxel map are updated. In this way, the frequency of calculating the dynamic and static attributes of voxels can be reduced, thereby reducing the amount of calculation. At the same time, considering that the dynamic and static attributes of voxels may also change with changes in the environment, the dynamic and static attributes of voxels can also be updated in real time, thereby improving the accuracy of point cloud dynamic and static segmentation.

[0079] The laser radar point cloud dynamic and static segmentation method provided in this embodiment can be applied to projects such as smart cities. For example, the laser radar can be set up at the intersection, and the traffic volume, pedestrian flow, etc. can be counted by analyzing the point cloud data collected by the laser radar. It can also be applied to other scenes based on the dynamic and static attributes of point clouds for analysis and identification. The laser radar point cloud dynamic and static segmentation method provided in this embodiment first voxelizes the target area, establishes a voxel map, and determines the dynamic and static attributes of each voxel in the voxel map, that is, divides all voxels into dynamic voxels and static voxels. Then, according to whether each data point in the point cloud frame falls into a dynamic voxel or a static voxel, it is determined whether each data point is a dynamic point or a static point. Among them, when determining the dynamic and static attributes of the voxel, for any voxel, the static and dynamic attributes of the voxel are preliminarily distinguished by the statistics of the point cloud frame in which the data point falls in the voxel. If most of the point cloud frames in the first number of frames have data points falling into a certain voxel, then the voxel tends to be a static voxel. For such voxels, the reflectivity can characterize the surface characteristics of the object. By counting the reflectivity of the data points falling into the voxel, it is determined whether the data points falling into the voxel between different point cloud frames belong to the same object. Specifically, the distribution of reflectivity can be calculated by Shannon entropy. If the entropy value is small, it means that the reflectivity is close to unity, and the data points falling into the voxel in different point cloud frames tend to belong to the same object, and the voxel is determined to be a static voxel; otherwise, they tend to belong to different objects, and the voxel is determined to be a dynamic voxel. Through this method, the probability of misjudging dynamic voxels as static voxels can be reduced. For example, if there is a constant stream of vehicles passing through the target area, for a certain voxel, many point cloud frames may have data points falling on this voxel, but the data points falling between different point cloud frames belong to different vehicles. If the object to which the data points falling into the voxel belong is not analyzed, there is a high probability that the voxel will be misjudged as a static voxel. By applying the method provided in the embodiment of the present application, this problem can be avoided, thereby improving the accuracy of dynamic and static segmentation of point clouds.

[0080] Further, as Figure 1 The specific implementation of the method shown in the embodiment provides a laser radar point cloud dynamic and static segmentation device, such as Figure 2 As shown, the device includes: a voxel map building module 31 and a voxel dynamic and static attribute determination module 32.

[0081] The voxel map building module 31 can be used to perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map;

[0082] The voxel static and dynamic attribute determination module 32 can be used to map data points in multiple point cloud frames to the voxel map, and calculate the static and dynamic attributes of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame.

[0083] In a specific application scenario, the data points in multiple point cloud frames are mapped to the voxel map, and before the dynamic and static properties of the target voxel are calculated based on the reflectivity of the data points mapped in each point cloud frame in the target voxel, the voxel dynamic and static property determination module 32 can be specifically used to map a preset first number of point cloud frames to the voxel map to obtain a mapping relationship between the data points in each point cloud frame and each voxel; for any voxel, according to the mapping relationship, the number of point cloud frames in which the number of data points mapped in any voxel is greater than zero is counted, recorded as the second frame number; the ratio of the second frame number to the first frame number is calculated, and if the ratio of the second frame number to the first frame number exceeds a preset first threshold, the voxel is marked as a target voxel.

[0084] In a specific application scenario, the dynamic and static attributes of the target voxel are calculated based on the reflectivity of the data points mapped in the target voxel of each point cloud frame. The voxel dynamic and static attribute determination module 32 can be specifically used to calculate the reflectivity of the data points mapped in the target voxel of the point cloud frame for each point cloud frame, and obtain multiple reflectivities corresponding to the target voxel; calculate the discrete eigenvalue of the target voxel based on the multiple reflectivities corresponding to the target voxel; and determine the dynamic and static attributes of the target voxel based on the discrete eigenvalue of the target voxel.

[0085] In a specific application scenario, the dynamic and static attributes of the target voxel are determined based on the discrete eigenvalue of the target voxel. The voxel dynamic and static attribute determination module 32 can be specifically used to determine that the target voxel is a static voxel when the discrete eigenvalue of the target voxel is less than a preset second threshold value, otherwise, determine that the target voxel is a dynamic voxel.

[0086] In a specific application scenario, the static and dynamic attributes of the target voxel are calculated according to the reflectivity of the data points mapped in the target voxel of each point cloud frame. The voxel static and dynamic attribute determination module 32 can be specifically used to calculate the discrete eigenvalue of the target voxel and the discrete eigenvalue of each bounding box voxel corresponding to the target voxel, wherein the discrete eigenvalue is calculated according to the reflectivity of the data points mapped in the voxel of each point cloud frame; the weight of the target voxel and multiple bounding box voxels are calculated according to the distance between each of the bounding box voxels and the target voxel; the weighted discrete eigenvalue of the target voxel is obtained according to the discrete eigenvalue of the target voxel and the weight of the target voxel, as well as the discrete eigenvalues ​​of multiple bounding box voxels and the weights of multiple bounding box voxels; when the weighted discrete eigenvalue is less than a preset second threshold, the target voxel is determined to be a static voxel, otherwise, the target voxel is determined to be a dynamic voxel.

[0087] In a specific application scenario, the device also includes a point cloud dynamic and static attribute determination module 33. After calculating the dynamic and static attributes of the target voxel, the point cloud dynamic and static attribute determination module 33 can be specifically used to map the data points in the point cloud frame to be segmented to the voxel map, and determine the dynamic and static attributes of the data points in the point cloud frame to be segmented based on the dynamic and static attributes of the voxels in the voxel map.

[0088] In a specific application scenario, the voxel dynamic and static attribute determination module 32 can also be used to reacquire multiple point cloud frames at a preset time interval, recalculate the dynamic and static attributes of the target voxel based on the reacquired multiple point cloud frames, and update the dynamic and static attributes of the target voxel in the voxel map.

[0089] It should be noted that for other corresponding descriptions of the functional units involved in the laser radar point cloud dynamic and static segmentation device provided in this embodiment, please refer to Figure 1 As well as the corresponding descriptions in the above embodiments, they will not be repeated here.

[0090] Based on the above Figure 1 As well as the method shown in the above embodiment, correspondingly, this embodiment also provides a storage medium on which a computer program is stored, and when the program is executed by the processor, the above Figure 1 And the laser radar point cloud dynamic and static segmentation method shown in the above embodiment.

[0091] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0092] Based on the above Figure 1 And the method shown in the above embodiment, and Figure 2 In order to achieve the above-mentioned purpose, the laser radar point cloud dynamic and static segmentation device embodiment shown in FIG. Figure 3 As shown, this embodiment also provides a computer device for laser radar point cloud dynamic and static segmentation, which can be a personal computer, a server, a smart phone, a tablet computer, a smart watch, or other network devices, etc. The computer device includes a storage medium and a processor; the storage medium is used to store computer programs and an operating system; the processor is used to execute the computer program to implement the above-mentioned Figure 1 And the methods shown in the above embodiments.

[0093] Optionally, the computer device may further include an internal memory, a communication interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, a display, an input device such as a keyboard, etc. Optionally, the communication interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0094] Those skilled in the art will understand that the computer device structure for dynamic and static segmentation of laser radar point cloud provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0095] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the above-mentioned computer device hardware and the software resources to be identified, and supports the operation of the information processing program and other software and / or programs to be identified. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing computer device.

[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, the target area is voxelized to establish a voxel map, and the dynamic and static attributes of the target voxels in the voxel map are first identified, and then the dynamic and static attributes of the data points in the point cloud frame can be determined according to the dynamic and static attributes of the target voxels. Among them, when identifying the dynamic and static attributes of the target voxels, the reflectivity of the data points mapped in the voxels of each point cloud frame is used to analyze whether the object corresponding to each point cloud frame in the voxel is the same object, so as to determine the dynamic and static attributes of the target voxels. Compared with the prior art, the method provided by the present application reduces the probability of dynamic voxels being misjudged as static voxels due to long-term occupancy, and can improve the accuracy of point cloud dynamic and static segmentation.

[0097] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0098] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A laser radar point cloud dynamic and static segmentation method, characterized in that: The method comprises: Perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map; The data points in a plurality of point cloud frames are mapped to the voxel map, and the dynamic and static properties of the target voxel are calculated based on the reflectivity of the data points mapped in the target voxel by each point cloud frame.

2. The method according to claim 1, characterized in that Before mapping the data points in the plurality of point cloud frames to the voxel map and calculating the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel by each point cloud frame, the method further comprises: Mapping a preset first number of point cloud frames to the voxel map to obtain a mapping relationship between a data point in each point cloud frame and each voxel; For any voxel, according to the mapping relationship, the number of the point cloud frames in which the number of data points mapped in the any voxel is greater than zero is counted, and recorded as a second frame number; The ratio of the second frame number to the first frame number is calculated, and if the ratio of the second frame number to the first frame number exceeds a preset first threshold, the voxel is marked as a target voxel.

3. The method according to claim 1 or 2, characterized in that: The step of calculating the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame includes: For each of the point cloud frames, calculating the reflectivity of the data points of the point cloud frame mapped in the target voxel, and obtaining a plurality of reflectivities corresponding to the target voxel; Calculating a discrete characteristic value of the target voxel according to a plurality of reflectivities corresponding to the target voxel; The dynamic and static attributes of the target voxel are determined based on the discrete feature values ​​of the target voxel.

4. The method according to claim 3, characterized in that The determining the dynamic and static attributes of the target voxel based on the discrete characteristic value of the target voxel comprises: When the discrete characteristic value of the target voxel is less than a preset second threshold, the target voxel is determined to be a static voxel; otherwise, the target voxel is determined to be a dynamic voxel.

5. The method according to claim 1 or 2, characterized in that: The step of calculating the dynamic and static properties of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame includes: Calculating a discrete eigenvalue of the target voxel and a discrete eigenvalue of each bounding box voxel corresponding to the target voxel, wherein the discrete eigenvalue is calculated based on the reflectivity of the data point mapped in the voxel by each point cloud frame; Calculating the weights of the target voxel and the plurality of bounding box voxels according to the distance between each of the bounding box voxels and the target voxel; Obtaining a weighted discrete eigenvalue of the target voxel according to the discrete eigenvalue of the target voxel and the weight of the target voxel, and the discrete eigenvalues ​​of a plurality of bounding box voxels and the weights of a plurality of bounding box voxels; When the weighted discrete eigenvalue is less than a preset second threshold, the target voxel is determined to be a static voxel; otherwise, the target voxel is determined to be a dynamic voxel.

6. The method according to claim 1, characterized in that After calculating the dynamic and static attributes of the target voxel, the method further includes: The data points in the point cloud frame to be segmented are mapped to the voxel map, and the dynamic and static properties of the data points in the point cloud frame to be segmented are determined according to the dynamic and static properties of the voxels in the voxel map.

7. The method according to claim 1, characterized in that The method further comprises: At preset time intervals, a plurality of the point cloud frames are reacquired, the dynamic and static properties of the target voxels are recalculated according to the reacquired plurality of the point cloud frames, and the dynamic and static properties of the target voxels in the voxel map are updated.

8. A laser radar point cloud dynamic and static segmentation device, characterized in that: The device comprises: A voxel map building module is used to perform voxel segmentation on the target area within the radar scanning range to obtain a voxel map; The voxel static and dynamic attribute determination module is used to map the data points in multiple point cloud frames to the voxel map, and calculate the static and dynamic attributes of the target voxel according to the reflectivity of the data points mapped in the target voxel of each point cloud frame.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images