Ground detection method, electronic device and storage medium

Abstract

The invention relates to a ground detection method, an electronic device and a storage medium, the electronic device performs rasterizing processing on point cloud data of a detection area to obtain rasterized point cloud data; determining a grid feature map based on the rasterized point cloud data, and determining a grid classification map and a height regression map according to the grid feature map; wherein the grid classification map is used for determining whether each grid is a ground grid; the height regression graph is used for determining height information of each grid from the ground; and performing ground detection processing based on the grid classification graph and the height regression graph to obtain a ground detection result corresponding to the detection area. Based on the scheme, the detection precision and the detection efficiency of ground detection can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a ground detection method, an electronic device and a storage medium. Background Art

[0002] In the application scenarios of autonomous driving perception, LiDAR has been increasingly used; related algorithms based on LiDAR have also become a hot topic in the industry; among the perception algorithms based on LiDAR, ground segmentation of laser point clouds is a typical task in laser perception. However, the current main ground detection algorithms mostly use algorithms based on traditional seed point extraction and plane fitting, which are insufficient in detection accuracy and speed. Summary of the invention

[0003] The embodiments of the present application provide a ground detection method, an electronic device, and a storage medium, which can effectively improve the detection accuracy and detection efficiency of ground detection.

[0004] In a first aspect, an embodiment of the present application provides a ground detection method, the method comprising:

[0005] Performing rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data;

[0006] Determine a grid feature map based on the rasterized point cloud data, and determine a grid classification map and a height regression map based on the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground;

[0007] Ground detection processing is performed based on the grid classification map and the height regression map to obtain the ground detection results corresponding to the detection area.

[0008] In this embodiment, after obtaining the point cloud data of the detection area, the electronic device can first rasterize the point cloud data, and then use the rasterized point cloud data to determine the raster feature map, and further use the raster feature map to determine the raster classification map that can be used to characterize whether each grid is classified as a ground grid, and the height regression map that can be used to determine the height information of each grid from the ground. The grid classification map and the height regression map can thus be used directly to complete the ground detection of the detection area, effectively determine the ground area in the detection area, and greatly improve the detection accuracy and efficiency of the ground detection.

[0009] Furthermore, in some embodiments, ground detection processing is performed based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area, including:

[0010] According to the coordinate information of the detection point, a first grid is determined in each grid of the grid classification map; wherein the detection point is any point in the point cloud data; and the first grid represents a grid in the grid classification map including a feature point corresponding to the detection point;

[0011] When the classification parameter of the first grid is greater than the first threshold, a second grid is determined in each grid of the height regression map according to the coordinate information of the detection point; wherein the second grid represents a grid including a feature point corresponding to the detection point in the height regression map;

[0012] When the height parameter of the second grid is less than or equal to the second threshold, the classification result of the detection point is determined to be a ground point, and the ground detection result is determined according to the classification results of each detection point in the point cloud data.

[0013] In this embodiment, when determining the ground detection result based on the grid classification map and the height regression map, for any detection point in the point cloud data, its first grid in the grid classification map can be determined first, that is, the grid including the feature point corresponding to the detection point in the grid classification map, and then the classification parameter of the first grid and the first threshold are compared. If the classification parameter of the first grid is greater than the first threshold, it means that the first grid is a ground grid. In order to improve the accuracy of ground detection, the second grid corresponding to the detection point can be further determined in the height regression map, and the height parameter of the second grid can be obtained. If the height parameter of the second grid is less than or equal to the second threshold, it can be determined that the detection point is a ground point, so that the ground detection result of the detection area can be determined according to the classification of whether each detection point is a ground point, thereby effectively improving the efficiency of ground detection.

[0014] Further, in some embodiments, after determining the first grid in each grid of the grid classification map according to the coordinate information of the detection point, the method further includes:

[0015] When the classification parameter of the first grid is less than or equal to the first threshold, the classification result of the detection point is determined to be a non-ground point.

[0016] In this embodiment, if the classification parameter of the first grid is less than or equal to the first threshold, the detection point can be determined to be a non-ground point, thereby enabling rapid distinction of non-ground areas in the detection area and improving ground detection efficiency.

[0017] Further, in some embodiments, when the classification parameter of the first grid is greater than the first threshold, after determining the second grid in each grid of the height regression map according to the coordinate information of the detection point, the method further includes:

[0018] When the height parameter of the second grid is greater than the second threshold, it is determined that the classification result of the detection point is a non-ground point.

[0019] In this embodiment, after the second grid is determined, if the height parameter of the second grid is greater than the second threshold, the detection point can also be determined to be a non-ground point, which can effectively improve the detection accuracy of the non-ground point.

[0020] Further, in some embodiments, determining a grid feature map based on the rasterized point cloud data includes:

[0021] Perform feature extraction on the rasterized point cloud data to obtain the grid feature information corresponding to each grid;

[0022] Perform feature transformation processing on the grid feature information based on a multi-layer perceptron to obtain transformed grid feature information;

[0023] A plane projection process is performed according to the transformed grid feature information to obtain a grid feature map.

[0024] In this embodiment, when determining the raster feature map corresponding to the rasterized point cloud data, the rasterized point cloud data can be first subjected to feature extraction processing, and then the raster feature information can be subjected to feature transformation processing using a multi-layer perceptron, and then a planar projection is performed based on the transformed raster feature information to obtain a raster feature map, which can effectively extract the feature information in the point cloud data and achieve more accurate feature expression.

[0025] Furthermore, in some embodiments, feature extraction processing is performed on the rasterized point cloud data to obtain grid feature information corresponding to each grid, including:

[0026] Perform plane projection processing on the rasterized point cloud data to obtain a raster plane map;

[0027] Determine the geometric center point of each grid in the grid plane;

[0028] Determine the centroid point of each grid according to the rasterized point cloud data;

[0029] The grid feature information is determined based on the coordinate information of the geometric center point, the centroid point, and the detection points within each grid.

[0030] In this embodiment, when determining the grid feature information, the rasterized point cloud data can be projected onto a plane to obtain a grid plane map, and the geometric center point of each grid can be determined in the grid plane map. At the same time, the rasterized point cloud data can be used to determine the centroid point of each grid, and then the grid feature information can be determined using the geometric center point and centroid point of each grid and the coordinate information of the detection points within each grid, thereby obtaining multi-dimensional grid feature information.

[0031] Further, in some embodiments, the coordinate information includes three-dimensional coordinate information corresponding to the detection point in the rasterized point cloud data, and two-dimensional coordinate information corresponding to the detection point in the grid plane; the grid feature information is determined based on the coordinate information of the geometric center point, the centroid point, and the detection point in each grid, including:

[0032] Determine first difference information according to the coordinate information of the centroid and the three-dimensional coordinate information;

[0033] Determine second difference information according to the coordinate information of the geometric center point and the two-dimensional coordinate information;

[0034] The grid feature information is determined based on the first difference information, the second difference information and the four-dimensional information of each detection point; wherein the four-dimensional information includes the three-dimensional coordinate information of the detection point in the point cloud data and the laser reflection intensity information.

[0035] In this embodiment, when determining the grid feature information, the first difference information from the detection point to the center of mass point of the grid and the three-dimensional coordinate information of the detection point in the grid can be used to determine the first difference information from the detection point to the center of mass point, and at the same time, the second difference information from the detection point to the geometric center point can be determined using the two-dimensional coordinate information of the geometric center point of the grid and the detection point in the grid in the grid plan view. Thus, based on the first difference information, the second difference information and the four-dimensional information of the detection point in the original point cloud data, multi-dimensional grid feature information is determined, the accuracy of the expression of the grid feature information is improved, and the reliability of subsequent ground detection based on the grid feature information is improved.

[0036] In a second aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a determination unit and a detection unit;

[0037] A determination unit is used to perform rasterization processing on the point cloud data corresponding to the detection area to obtain rasterized point cloud data; and determine a raster feature map based on the rasterized point cloud data, and determine a raster classification map and a height regression map according to the raster feature map; wherein the raster classification map is used to determine whether each raster is a ground raster; and the height regression map is used to determine the height information of each raster from the ground;

[0038] The detection unit is used to perform ground detection processing based on the grid classification map and the height regression map to obtain the ground detection result corresponding to the detection area.

[0039] In this embodiment, after obtaining the point cloud data of the detection area, the electronic device can first rasterize the point cloud data, and then use the rasterized point cloud data to determine the raster feature map, and further use the raster feature map to determine the raster classification map that can be used to characterize whether each grid is classified as a ground grid, and the height regression map that can be used to determine the height information of each grid from the ground. The grid classification map and the height regression map can thus be used directly to complete the ground detection of the detection area, effectively determine the ground area in the detection area, and greatly improve the detection accuracy and efficiency of the ground detection.

[0040] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory storing instructions executable by the processor; when the instructions are executed by the processor, the above-mentioned ground detection method is implemented.

[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned ground detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the implementation process of the ground detection method provided in the embodiment of the present application Figure 1 ;

[0043] Figure 2 Schematic diagram of the implementation process of the ground detection method provided in the embodiment of the present application Figure 2 ;

[0044] Figure 3 Schematic diagram of the implementation process of the ground detection method provided in the embodiment of the present application Figure 3 ;

[0045] Figure 4 Schematic diagram of the implementation process of the ground detection method provided in the embodiment of the present application Figure 4 ;

[0046] Figure 5 Schematic diagram of the composition structure of the electronic device provided in the embodiment of the present application Figure 1 ;

[0047] Figure 6 Schematic diagram of the composition structure of the electronic device provided in the embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0048] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0049] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0050] In the application scenarios of autonomous driving perception, LiDAR has been increasingly used; related algorithms based on LiDAR have also become a hot topic in the industry; in the perception algorithms based on LiDAR, ground segmentation of laser point cloud is a typical task in laser perception. However, the current main ground detection algorithms mostly use algorithms based on traditional seed point extraction plus plane fitting, which are insufficient in detection accuracy, speed and reliability; for example, most ground detection algorithms are based on traditional algorithms, which have great challenges in dealing with the robustness of different ground surfaces. When the ground is curved or the visibility of the ground point cloud is poor, these algorithms usually perform poorly; and some ground detection algorithms are extremely complex, resulting in excessive algorithm latency, making it difficult to run in real time on devices with limited computing power; and most common ground detection algorithms are based on plane fitting to represent the ground, which has poor accuracy when the ground is irregular, and cannot simultaneously achieve the segmentation of ground point clouds and the accurate expression of ground height; in addition, although some ground segmentation algorithms based on deep learning have relatively good performance, their network structure is often relatively complex, resulting in unnecessary performance waste, and cannot be deployed on devices with insufficient computing power.

[0051] In order to solve the above problems, in the embodiments of the present application, a ground detection method, an electronic device and a storage medium are proposed. The electronic device can perform rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data; determine the raster feature map based on the rasterized point cloud data, and determine the raster classification map and the height regression map based on the raster feature map; wherein the raster classification map is used to determine whether each grid is a ground grid; the height regression map is used to determine the height information of each grid from the ground; perform ground detection processing based on the raster classification map and the height regression map to obtain the ground detection result corresponding to the detection area. Based on the above scheme, the detection accuracy, detection efficiency and reliability of ground detection can be effectively improved.

[0052] The technical solutions 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.

[0053] Figure 1 Schematic diagram of the implementation process of the ground detection method provided in the embodiment of the present application Figure 1 ,like Figure 1 As shown, in an embodiment of the present application, a ground detection method of an electronic device may include the following steps:

[0054] Step 101: rasterize the point cloud data of the detection area to obtain rasterized point cloud data.

[0055] In an embodiment of the present application, the electronic device may first perform rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data.

[0056] It should be noted that in the embodiments of the present application, the electronic device can be any device with communication and storage functions, and the present application does not limit it; for example, the electronic device can be an in-vehicle electronic device, a car machine, a personal computer (PC), a laptop computer and other electronic devices.

[0057] In an embodiment of the present application, the detection area may be any area, and the detection area may be scanned by a laser radar to obtain point cloud data of the detection area.

[0058] Exemplarily, each point in the point cloud data can be represented by four-dimensional information (x, y, z, intensity), where intensity represents the laser reflection intensity, and x, y, z represent three-dimensional coordinate information; when performing rasterization processing, the point cloud data can be rasterized according to a preset grid size, for example, the preset grid size is 1 meter × 1 meter. After the point cloud data is rasterized, each grid in the obtained rasterized point cloud data can include a maximum of N points. If the number of points falling in this grid is less than N, the grid is filled with points with a value of 0. If the number of points falling in this grid is greater than N, N points are randomly sampled as points in the grid, and the points in the grid do not need to be sorted.

[0059] Step 102, determine a grid feature map based on the rasterized point cloud data, and determine a grid classification map and a height regression map based on the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground.

[0060] In an embodiment of the present application, the electronic device can rasterize the point cloud data of the detection area to obtain the rasterized point cloud data, determine the raster feature map based on the rasterized point cloud data, and determine the raster classification map and the height regression map based on the raster feature map; wherein the raster classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground.

[0061] In an embodiment of the present application, the grid feature map may be a two-dimensional feature map constructed based on feature information extracted from rasterized point cloud data.

[0062] In an embodiment of the present application, the grid feature map can be input into a trained convolutional neural network model to obtain a grid classification map and a height regression map.

[0063] In some embodiments of the present application, after inputting the raster feature map into the trained convolutional neural network model, in addition to obtaining the raster classification map and the height regression map, the ground plane equation can also be obtained. The ground plane equation can play an auxiliary supervision role to improve the accuracy of the trained convolutional neural network model in the process of calculating the output raster classification map and the height regression map.

[0064] In some embodiments of the present application, when the electronic device determines a raster feature map based on rasterized point cloud data, it can perform feature extraction processing on the rasterized point cloud data to obtain raster feature information corresponding to each grid; then perform feature transformation processing on the raster feature information based on a multi-layer perceptron to obtain transformed raster feature information; then perform plane projection processing based on the transformed raster feature information to obtain a raster feature map.

[0065] In an embodiment of the present application, the grid feature information may include first difference information, second difference information, and four-dimensional information of each detection point in the point cloud data and the detection point in the original point cloud data.

[0066] In some embodiments of the present application, when the electronic device performs feature extraction processing on the rasterized point cloud data to obtain grid feature information corresponding to each grid, it can perform plane projection processing on the rasterized point cloud data to obtain a grid plane map; then determine the geometric center point of each grid in the grid plane map; determine the centroid point of each grid based on the rasterized point cloud data; and thereby determine the grid feature information based on the coordinate information of the geometric center point, the centroid point, and the detection points within each grid.

[0067] In some embodiments of the present application, the coordinate information may include three-dimensional coordinate information corresponding to the detection point in the rasterized point cloud data, and two-dimensional coordinate information corresponding to the detection point in the grid plane.

[0068] In some embodiments of the present application, when the electronic device determines the grid feature information based on the coordinate information of the geometric center point, the centroid point, and the detection points within each grid, it can determine the first difference information based on the coordinate information of the centroid point and the three-dimensional coordinate information; determine the second difference information based on the coordinate information of the geometric center point and the two-dimensional coordinate information; and thereby determine the grid feature information based on the first difference information, the second difference information, and the four-dimensional information of each detection point; wherein the four-dimensional information includes the three-dimensional coordinate information and laser reflection intensity information of the detection point in the point cloud data.

[0069] In some embodiments of the present application, the first difference information may include a first difference, a second difference and a third difference; the second difference information may include a fourth difference and a fifth difference; wherein the first difference represents the difference between the coordinate value of the x-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the x-axis in the coordinate information of the center of mass, the second difference represents the difference between the coordinate value of the y-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the y-axis in the coordinate information of the center of mass, the third difference represents the difference between the coordinate value of the z-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the z-axis in the coordinate information of the center of mass; the fourth difference represents the difference between the coordinate value of the x-axis in the two-dimensional coordinate information of the detection point and the coordinate value of the x-axis in the coordinate information of the geometric center point; the fifth difference represents the difference between the coordinate value of the y-axis in the two-dimensional coordinate information of the detection point and the coordinate value of the y-axis in the coordinate information of the geometric center point.

[0070] Exemplarily, the rasterized point cloud data includes P grids, each grid contains N points; for a certain grid, the three-dimensional coordinate information of the detection point included therein can be expressed as (Xi, Yi, Zi), and the coordinates of the centroid point of the grid can be expressed as (Xc, Yc, Zc), then the coordinate value of the x-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the x-axis in the coordinates of the centroid point can be subtracted to obtain a first difference Xi-Xc, and at the same time, the coordinate value of the y-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the y-axis in the coordinates of the centroid point are subtracted to obtain a second difference Yi-Yc, and, the coordinate value of the z-axis in the three-dimensional coordinate information of the detection point and the coordinate value of the z-axis in the coordinates of the centroid point are subtracted to obtain a third difference Zi-Zc, thereby obtaining the first difference information of the detection point based on the above-mentioned first difference Xi-Xc, second difference Yi-Yc and third difference Zi-Zc.

[0071] Exemplarily, the rasterized point cloud data is subjected to a planar projection process to obtain a grid plane view from a top view perspective. In the grid plane view, each grid may include a geometric center point. For example, the coordinates of the geometric center point of a certain grid are expressed as (Xg, Yg). In the grid plane view, the two-dimensional coordinate information of any detection point in the grid is expressed as (Xi, Yi). Then, the coordinate value of the x-axis in the two-dimensional coordinate information of the detection point and the coordinate value of the x-axis in the coordinates of the geometric center point can be subtracted to obtain a fourth difference Xi-Xg. At the same time, the coordinate value of the y-axis in the two-dimensional coordinate information of the detection point and the coordinate value of the y-axis in the coordinates of the geometric center point can be subtracted to obtain a fifth difference Yi-Yg. Thus, the second difference information of the detection point can be obtained according to the above fourth difference Xi-Xg and fifth difference Yi-Yg.

[0072] Exemplarily, for the detection point, the raster feature information composed of the three-dimensional information in the first difference information, the two-dimensional information in the second difference information, and the four-dimensional information of the detection point in the original point cloud data can contain information in nine dimensions. That is, through feature extraction processing, the original four-dimensional information can be expanded to nine dimensions to enrich the expression of feature information, thereby facilitating improving the accuracy of subsequent ground detection.

[0073] In some embodiments of the present application, after the grid feature information is subjected to feature transformation processing based on a multi-layer perceptron, the dimension of the transformed grid feature information is not limited in the present application and may be the same as or different from the dimension of the grid feature information; for example, the grid feature information includes information of 9 dimensions. After the grid feature information is subjected to feature transformation processing based on a multi-layer perceptron, the feature dimension may become C, and C may be greater than 9, equal to 9, or less than 9; after the feature transformation processing, the feature information may be further optimized so that the transformed grid feature information is obtained.

[0074] Exemplarily, the transformed grid feature information includes the transformed grid feature information corresponding to P grids, and the transformed grid feature information corresponding to the P grids is projected onto a plane, that is, the P grids are projected onto a plane to obtain a two-dimensional grid feature map from a top view perspective. The relevant parameter information of the grid feature map can be expressed as (C, Yn, Xn), where C represents the dimension of the feature information of each detection point, Yn represents the number of grids in the y-axis direction of the grid feature map, and Xn represents the number of grids in the x-axis direction of the grid feature map.

[0075] In some embodiments of the present application, the grid classification map may include probability information of each grid being classified as a ground grid; the height regression map may include height reference information of the detection point in each grid from the ground.

[0076] In some embodiments of the present application, the initial convolutional neural network model can be trained to obtain a trained convolutional neural network model; during training, three loss functions can be designed for the parameters of the ground plane equation, the height regression map, and the grid classification map output by the model, respectively. For example, the absolute error loss (L1 loss) can be used to calculate the loss function value corresponding to the parameters of the ground plane equation output by the model, the Huber loss can be used to calculate the loss function value corresponding to the height regression map output by the model, and the binary cross-entropy loss (Binary Cross-Loss, BCE loss) can be used to calculate the loss function value corresponding to the grid classification map output by the model. The total loss function value can be obtained by weighted calculation based on the above three loss function values, and the initial convolutional neural network model can be updated using the total loss function value to obtain the trained convolutional neural network model.

[0077] In some embodiments of the present application, before the raster feature map is input into the trained convolutional neural network model, the raster feature map can be subjected to data enhancement processing, and the data enhancement processing can include at least one of global rotation, z-value translation, and laser point cloud noise simulation; wherein, global rotation can be a rotation of the raster feature map in any direction, z-value translation can be a translation of the raster feature map in the direction of the z-axis, and laser point cloud noise simulation can be adding noise to the raster feature map; through data enhancement processing, the interference of the input raster feature map to the trained convolutional neural network model can be increased to improve the performance of the trained convolutional neural network model.

[0078] Step 103: Perform ground detection processing based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area.

[0079] In an embodiment of the present application, the electronic device can determine a grid feature map based on the rasterized point cloud data, and after determining a grid classification map and a height regression map based on the grid feature map, perform ground detection processing based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area.

[0080] In some embodiments of the present application, the ground detection result may represent the result of segmenting the ground area and the non-ground area in the detection area.

[0081] In some embodiments of the present application, when the electronic device performs ground detection processing based on the grid classification map and the height regression map to obtain the ground detection result corresponding to the detection area, the first grid can be determined in each grid of the grid classification map according to the coordinate information of the detection point; wherein the detection point is any point in the point cloud data; the first grid represents the grid in the grid classification map that includes the feature point corresponding to the detection point; and then, when the classification parameter of the first grid is greater than the first threshold, the second grid is determined in each grid of the height regression map according to the coordinate information of the detection point; wherein the second grid represents the grid that includes the feature point corresponding to the detection point in the height regression map; when the height parameter of the second grid is less than or equal to the second threshold, the classification result of the detection point is determined to be a ground point, and the ground detection result is determined according to the classification results of each detection point in the point cloud data.

[0082] Exemplarily, take any detection point in the point cloud data, and determine the feature point corresponding to the detection point according to its coordinates in the feature points in each grid of the grid classification map, so that the grid where the feature point is located, that is, the first grid, can be determined; based on the same method, when it is necessary to determine the second grid corresponding to the detection point, the feature point corresponding to the detection point can be determined according to the coordinates of the detection point in the feature points in each grid of the height regression map, so that the grid where the feature point is located can be determined as the second grid.

[0083] In an embodiment of the present application, the classification parameter may be probability information of a grid being classified as a ground grid; when the classification parameter is greater than a first threshold, the grid may be considered to contain ground points, otherwise, the grid does not contain ground points.

[0084] In the embodiments of the present application, the specific value of the first threshold is not limited in the present application. For example, the first threshold may be 0.1. When the classification parameter is greater than 0.1, it can be determined that the first grid contains ground points.

[0085] In some embodiments of the present application, when it is determined that the classification parameter of the first grid is greater than the first threshold, it will continue to determine that the detection point corresponds to the second grid in the height regression map, so as to accurately judge whether the detection point is a ground point based on the height parameter of the second grid, which can effectively improve the detection accuracy of ground detection.

[0086] In the embodiments of the present application, the specific value of the second threshold is not limited in the present application.

[0087] Exemplarily, the second threshold is represented as Zgnd; when it is determined that the classification parameter of the first grid corresponding to a detection point is greater than 0.1, the second grid corresponding to the detection point can be further determined in the height regression map, and the height parameter of the second grid can be obtained. If the height parameter of the second grid is less than or equal to Zgnd, the detection point can be determined as a ground point, otherwise it is a non-ground point.

[0088] In some embodiments of the present application, after the electronic device determines the first grid among the grids of the grid classification map, it may also determine that the classification result of the detection point is a non-ground point when the classification parameter of the first grid is less than or equal to a first threshold.

[0089] In some embodiments of the present application, when the classification parameter of the first grid is greater than the first threshold, the electronic device may determine the second grid in each grid of the height regression map based on the coordinate information of the detection point, and may also determine that the classification result of the detection point is a non-ground point when the height parameter of the second grid is greater than the second threshold.

[0090] The embodiment of the present application provides a ground detection method, wherein the electronic device performs rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data; determines a raster feature map based on the rasterized point cloud data, and determines a raster classification map and a height regression map based on the raster feature map; wherein the raster classification map is used to determine whether each grid is a ground grid; the height regression map is used to determine the height information of each grid from the ground; performs ground detection processing based on the raster classification map and the height regression map to obtain a ground detection result corresponding to the detection area. It can be seen that after obtaining the point cloud data of the detection area, the electronic device can first rasterize the point cloud data, and then use the rasterized point cloud data to determine the raster feature map, and further use the raster feature map to determine the raster classification map that can be used to characterize whether each grid is classified as a ground grid, and the height regression map that can be used to determine the height information of each grid from the ground, so that the ground detection of the detection area can be completed directly using the raster classification map and the height regression map, effectively determining the ground area in the detection area, and greatly improving the detection accuracy and detection efficiency of the ground detection.

[0091] Based on the above embodiments, in another embodiment of the present application, exemplarily, the ground detection method of the present application can be mainly divided into three modules, namely, a point cloud feature extraction module, a convolutional neural network module and a ground point post-processing module; after obtaining the point cloud data, the electronic device can input the point cloud data into the point cloud feature extraction module, and then after constructing a grid feature map based on the extracted relevant features, the grid feature map is input into the convolutional neural network module to obtain a height regression map and a grid classification map, and finally, the ground point post-processing module is used to perform ground detection processing on the height regression map and the grid classification map, and the ground detection result is input.

[0092] For example, Figure 2As shown, when using point cloud data to determine a grid feature map, the point cloud data can be firstly rasterized to obtain rasterized point cloud data (step 201). For example, the rasterized point cloud data can be P grids containing a fixed number of detection points. Feature extraction is then performed on the rasterized point cloud data to obtain grid feature information (step 202). The grid feature information is then transformed using a multi-layer perceptron to obtain transformed grid feature information (step 203). Finally, the transformed grid feature information is plane projected to obtain a grid feature map (step 204).

[0093] In some embodiments of the present application, the input point cloud can first be rasterized. Usually, each point in the original point cloud data can contain information in four dimensions, namely (x, y, z, intensity); this embodiment adopts point cloud rasterization processing under a top view, and the preset grid size is 1 meter × 1 meter. The preset grid size can be adjusted according to actual needs, and this application does not limit it; after rasterization processing, each grid contains a maximum of N points. If the number of points falling in this grid is less than N, the points with a value of 0 are filled. If the number of points falling in this grid is greater than N, N points are randomly sampled and extracted. The points in the grid do not need to be sorted.

[0094] In some embodiments of the present application, it is assumed that after the above-mentioned rasterization processing, P grids are obtained, each grid contains N points, and the current dimension of each point is 4 dimensions; the point cloud feature extraction module can further extract feature information in each grid, including: the first difference information from the detection point (Xi, Yi, Zi) in the grid to the centroid point (Xc, Yc, Zc) of the grid; and the second difference information from the detection point (Xi, Yi) in the grid to the geometric center (Xg, Yg) of the grid in the grid plane. Through the above simple feature extraction, the features of each point are expanded from the original four dimensions (x, y, z, intensity) to 9 dimensions.

[0095] In some embodiments of the present application, the point cloud feature extraction module can use a multi-layer perceptron to perform feature transformation processing on the basis of the 9-dimensional grid feature information obtained above, for example, the feature dimension after transformation is C; after the multi-layer perceptron performs feature transformation on the feature information corresponding to all grids, P new grids of C dimensions can be obtained.

[0096] In some embodiments of the present application, planar projection processing can also be performed on P new grids of C dimensions to obtain a grid feature map from a top-down perspective. For example, relevant parameter information of the grid feature map can be expressed as (C, Yn, Xn), where C represents the dimension of the feature information of each detection point, Yn represents the number of grids in the y-axis direction of the grid feature map, and Xn represents the number of grids in the x-axis direction of the grid feature map; the grid feature map can be used as input for a subsequent convolutional neural network model.

[0097] In some embodiments of the present application, the trained convolutional neural network model may include different convolution kernels; for example, Figure 3 As shown, after the grid feature map is input into the trained convolutional neural network model (step 205), the parameters a, b, and d of the ground plane equation can be obtained through three layers of convolution operations and downsampling operations (steps 206, 207, 208, and 209), respectively; wherein, after the input feature map undergoes the first layer of convolution operations and downsampling operations (step 206), a feature map with a dimension of (2C, Yn / 2, Xn / 2) can be obtained, after the second layer of convolution operations and downsampling operations (step 207), a feature map with a dimension of (4C, Yn / 4, Xn / 4) can be obtained, and after the third layer of convolution operations and downsampling operations (step 208), a feature map with a dimension of (8C, Yn / 8, Xn / 8) can be obtained; wherein, the ground plane equation can be expressed as ax+by+z+d=0, and the trained convolutional neural network model can obtain the three parameters a, b, and d in the ground plane equation according to the grid feature map; from Figure 3 It can be seen that the output after the first layer of convolution and downsampling operation can be used to merge the output after the second layer of convolution and downsampling operation, merging operation (step 211) and convolution and upsampling operation (step 212) (step 210), and then the merged result is convolved and upsampled (step 213), and then the convolution operation (step 214 and step 215) is performed again to obtain the grid classification map (step 216) and the height regression map (step 217), wherein the merging operation of the second layer is completed based on the output of the third layer of convolution and downsampling operation, merging operation (step 218) and convolution and upsampling operation (step 219).

[0098] In some embodiments of the present application, the height regression map can be used to represent the height of each grid from the ground in a top view, and the grid classification map can be used to determine whether each grid is a ground grid in a top view.

[0099] In some embodiments of the present application, the loss function used by the convolutional neural network model during the training process can also be divided into three parts, and three loss functions can be designed for the parameters of the ground plane equation, the height regression map, and the grid classification map output by the model respectively; for example, the L1 loss can be used to calculate the loss function value corresponding to the parameters of the ground plane equation output by the model, the Huber loss can be used to calculate the loss function value corresponding to the height regression map output by the model, and the BCE loss can be used to calculate the loss function value corresponding to the grid classification map output by the model, and the total loss function value can be obtained by weighted calculation based on the above three loss function values, so as to use the total loss function value to update the initial convolutional neural network model to obtain the trained convolutional neural network model.

[0100] In some embodiments of the present application, before the raster feature map is input into the trained convolutional neural network model, the raster feature map can be subjected to data enhancement processing, and the data enhancement processing may include at least one of global rotation, z-value translation, and laser point cloud noise simulation.

[0101] In some embodiments of the present application, the ground point post-processing module can perform ground point and non-ground point segmentation based on the height regression map and the ground classification map output by the trained convolutional neural network module; for example, the dimensions of the two feature maps, the height regression map and the ground classification map, are both Yn×Xn, and the ground classification map can represent the probability p that each grid is a ground grid, that is, the ground classification map can include Yn×Xn probabilities p. Assuming the first threshold is 0.1, when p>0.1, it means that this grid contains ground points and can be regarded as a ground grid, otherwise it does not contain ground points; the height regression map includes the estimated ground height of each grid, that is, the height parameter; assuming the second threshold is Zgnd, if the height parameter z corresponding to a grid>Zgnd, the points in the grid can be segmented as non-ground points, otherwise they are ground points.

[0102] In some embodiments of the present application, if the first grid cannot be determined in the grids of the grid classification map based on the coordinate information of the detection point, that is, there is no grid corresponding to the detection point in the grid classification map, then it can be determined that the detection point is not within the valid range of the rasterization setting, and the detection point can be directly classified as a non-ground point.

[0103] In some embodiments of the present application, if the grid to which a detection point belongs is determined to be a non-ground grid according to the ground classification map, that is, the grid does not contain ground points, all points in the grid can be classified as non-ground points.

[0104] For example, Figure 4As shown, a detection point can be first taken out from the point cloud data (step 301), and then the grid corresponding to the detection point is determined in the grid classification map (step 302), and it is determined whether it is a ground grid (step 303). If not, it can be determined as a non-ground point (step 304), and the next detection point is taken for ground detection; if it is, the grid corresponding to the detection point is determined in the height regression map (step 305), and then it is determined whether the height parameter is less than or equal to Zgnd (step 306). If so, it can be determined as a ground point (step 307), if not, it is determined as a non-ground point; wherein, when determining whether the grid is classified as a ground grid, the classification parameter of the grid can be compared with the first threshold value. If it is greater than the first threshold value, it is explained as a ground grid, otherwise it is not a ground grid.

[0105] In summary, this application divides the ground detection task of lidar point cloud into two tasks: grid semantic segmentation and ground height regression. The outputs of the two tasks are combined to achieve efficient and accurate ground detection without plane fitting. While accurately estimating the ground surface, it can also directly output the relevant parameters of the ground plane equation. This solution can be directly based on the trained convolutional neural network and has good robustness for point clouds in different scenes. The convolutional neural network has a simple structure and low computing power requirements and can be deployed on embedded devices. A simple ground point post-processing module can be used to quickly segment the ground and non-ground point clouds.

[0106] The embodiment of the present application provides a ground detection method, wherein the electronic device performs rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data; determines a raster feature map based on the rasterized point cloud data, and determines a raster classification map and a height regression map based on the raster feature map; wherein the raster classification map is used to determine whether each grid is a ground grid; the height regression map is used to determine the height information of each grid from the ground; performs ground detection processing based on the raster classification map and the height regression map to obtain a ground detection result corresponding to the detection area. It can be seen that after obtaining the point cloud data of the detection area, the electronic device can first rasterize the point cloud data, and then use the rasterized point cloud data to determine the raster feature map, and further use the raster feature map to determine the raster classification map that can be used to characterize whether each grid is classified as a ground grid, and the height regression map that can be used to determine the height information of each grid from the ground, so that the ground detection of the detection area can be completed directly using the raster classification map and the height regression map, effectively determining the ground area in the detection area, and greatly improving the detection accuracy and detection efficiency of the ground detection.

[0107] Based on the above embodiment, in another embodiment of the present application, Figure 5 Schematic diagram of the composition structure of the electronic device provided in the embodiment of the present application Figure 1 ,like Figure 5 As shown, the electronic device 1 provided in the embodiment of the present application may include a determination unit 11 and a detection unit 12 .

[0108] The determination unit 11 can be used to rasterize the point cloud data corresponding to the detection area to obtain rasterized point cloud data; and determine the raster feature map based on the rasterized point cloud data, and determine the raster classification map and the height regression map according to the raster feature map; wherein the raster classification map is used to determine whether each grid is a ground grid; the height regression map is used to determine the height information of each grid from the ground.

[0109] The detection unit 12 can be used to perform ground detection processing based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area.

[0110] In some embodiments of the present application, the detection unit 12 can also be used to determine a first grid in each grid of the grid classification map based on the coordinate information of the detection point; wherein the detection point is any point in the point cloud data; the first grid represents the grid in the grid classification map including the feature point corresponding to the detection point; and when the classification parameter of the first grid is greater than the first threshold, determine the second grid in each grid of the height regression map based on the coordinate information of the detection point; wherein the second grid represents the grid in the height regression map including the feature point corresponding to the detection point; and when the height parameter of the second grid is less than or equal to the second threshold, determine that the classification result of the detection point is a ground point, and determine the ground detection result based on the classification results of each detection point in the point cloud data.

[0111] In some embodiments of the present application, the detection unit 12 can also be used to determine the classification result of the detection point as a non-ground point after determining the first grid in each grid of the grid classification map based on the coordinate information of the detection point, when the classification parameter of the first grid is less than or equal to a first threshold.

[0112] In some embodiments of the present application, the detection unit 12 can also be used to determine the second grid in each grid of the height regression map based on the coordinate information of the detection point when the classification parameter of the first grid is greater than the first threshold, and determine that the classification result of the detection point is a non-ground point when the height parameter of the second grid is greater than the second threshold.

[0113] In some embodiments of the present application, the determination unit 11 can also be used to perform feature extraction processing on the rasterized point cloud data to obtain raster feature information corresponding to each grid; and perform feature transformation processing on the raster feature information based on a multi-layer perceptron to obtain transformed raster feature information; and perform plane projection processing based on the transformed raster feature information to obtain a raster feature map.

[0114] In some embodiments of the present application, the determination unit 11 can also be used to perform planar projection processing on the rasterized point cloud data to obtain a grid plane map; and determine the geometric center point of each grid in the grid plane map; and determine the centroid point of each grid based on the rasterized point cloud data; and determine the grid feature information based on the coordinate information of the geometric center point, the centroid point and the detection points in each grid.

[0115] In some embodiments of the present application, the coordinate information includes three-dimensional coordinate information corresponding to the detection point in the rasterized point cloud data, and two-dimensional coordinate information corresponding to the detection point in the grid plane; the determination unit 11 can also be used to determine the first difference information based on the coordinate information and the three-dimensional coordinate information of the centroid point; and determine the second difference information based on the coordinate information and the two-dimensional coordinate information of the geometric center point; and determine the grid feature information based on the first difference information, the second difference information and the four-dimensional information of each detection point; wherein the four-dimensional information includes the three-dimensional coordinate information and the laser reflection intensity information of the detection point in the point cloud data.

[0116] In the embodiments of the present application, further, Figure 6 Schematic diagram of the composition structure of the electronic device provided in the embodiment of the present application Figure 2 ,like Figure 6 As shown, the electronic device 1 provided in the embodiment of the present application may further include a processor 13 and a memory 14 storing executable instructions of the processor 13 ; further, the electronic device 1 may further include a communication interface 15 and a bus 16 for connecting the processor 13 , the memory 14 and the communication interface 15 .

[0117] In the embodiment of the present application, the processor 13 can be at least one of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), a controller, a microcontroller, and a microprocessor. It can be understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other, and the embodiment of the present application is not specifically limited. The electronic device 10 can also include a memory 14, which can be connected to the processor 13, wherein the memory 14 is used to store executable program code, the program code includes computer operation instructions, and the memory 14 may include a high-speed RAM memory, and may also include a non-volatile memory, for example, at least two disk memories.

[0118] In the embodiment of the present application, the bus 16 is used to connect the communication interface 15, the processor 13 and the memory 14, and the mutual communication between these devices.

[0119] In the embodiment of the present application, the memory 14 is used to store instructions and data.

[0120] Further, in an embodiment of the present application, the processor 13 is used to perform rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data;

[0121] Determine a grid feature map based on the rasterized point cloud data, and determine a grid classification map and a height regression map based on the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground;

[0122] Ground detection processing is performed based on the grid classification map and the height regression map to obtain the ground detection results corresponding to the detection area.

[0123] In practical applications, the memory 14 may be a volatile memory, such as a random access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 13.

[0124] In addition, each functional module in this embodiment can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or software functional modules.

[0125] If the integrated unit is implemented in the form of a software function module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform all or part of the steps of the method of this embodiment. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk, etc. Various media that can store program codes.

[0126] The embodiment of the present application provides an electronic device, which includes a determination unit and a detection unit; the determination unit is used to rasterize the point cloud data corresponding to the detection area to obtain rasterized point cloud data; and determine the grid feature map based on the rasterized point cloud data, and determine the grid classification map and the height regression map according to the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; the height regression map is used to determine the height information of each grid from the ground; the detection unit is used to perform ground detection processing based on the grid classification map and the height regression map to obtain the ground detection result corresponding to the detection area. It can be seen that after obtaining the point cloud data of the detection area, the electronic device can first rasterize the point cloud data, and then use the rasterized point cloud data to determine the grid feature map, and further use the grid feature map to determine the grid classification map that can be used to characterize whether each grid is classified as a ground grid, and the height regression map that can be used to determine the height information of each grid from the ground, so that the ground detection of the detection area can be completed directly using the grid classification map and the height regression map, effectively determining the ground area in the detection area, and greatly improving the detection accuracy and detection efficiency of the ground detection.

[0127] Specifically, the program instructions corresponding to a ground detection method in this embodiment can be stored in a storage medium such as a CD, a hard disk, a USB flash drive, etc. When the program instructions corresponding to a ground detection method in the storage medium are read or executed by an electronic device, the following steps are included:

[0128] Performing rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data;

[0129] Determine a grid feature map based on the rasterized point cloud data, and determine a grid classification map and a height regression map based on the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground;

[0130] Ground detection processing is performed based on the grid classification map and the height regression map to obtain the ground detection results corresponding to the detection area.

[0131] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0132] The present application is described with reference to implementation flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0133] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which is implemented in the implementation flow diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process described in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0135] The above embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Any equivalent substitution or change made by a person skilled in the art based on the present invention is within the protection scope of the present invention.

Claims

1. A ground detection method, characterized in that: The method comprises: Performing rasterization processing on the point cloud data of the detection area to obtain rasterized point cloud data; Determine a grid feature map based on the rasterized point cloud data, and determine a grid classification map and a height regression map based on the grid feature map; wherein the grid classification map is used to determine whether each grid is a ground grid; and the height regression map is used to determine the height information of each grid from the ground; Ground detection processing is performed based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area.

2. The method according to claim 1, characterized in that The performing ground detection processing based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area includes: According to the coordinate information of the detection point, a first grid is determined in each grid of the grid classification map; wherein the detection point is any point in the point cloud data; and the first grid represents a grid in the grid classification map including a feature point corresponding to the detection point; When the classification parameter of the first grid is greater than a first threshold, determining a second grid in each grid of the height regression map according to the coordinate information of the detection point; wherein the second grid represents a grid in the height regression map including a feature point corresponding to the detection point; When the height parameter of the second grid is less than or equal to the second threshold, the classification result of the detection point is determined to be a ground point, and the ground detection result is determined according to the classification results of each of the detection points in the point cloud data.

3. The method according to claim 2, characterized in that After determining the first grid in each grid of the grid classification map according to the coordinate information of the detection point, the method further includes: When the classification parameter of the first grid is less than or equal to the first threshold, the classification result of the detection point is determined to be a non-ground point.

4. The method according to claim 3, characterized in that In the case where the classification parameter of the first grid is greater than the first threshold, after determining the second grid in each grid of the height regression map according to the coordinate information of the detection point, the method further includes: When the height parameter of the second grid is greater than the second threshold, it is determined that the classification result of the detection point is a non-ground point.

5. The method according to any one of claims 1 to 4, characterized in that: The step of determining a grid feature map based on the rasterized point cloud data comprises: Performing feature extraction processing on the rasterized point cloud data to obtain grid feature information corresponding to each grid; Performing feature transformation processing on the grid feature information based on a multi-layer perceptron to obtain transformed grid feature information; Plane projection processing is performed according to the transformed grid feature information to obtain the grid feature map.

6. The method according to claim 5, characterized in that The performing feature extraction processing on the rasterized point cloud data to obtain grid feature information corresponding to each grid includes: Performing plane projection processing on the rasterized point cloud data to obtain a raster plane map; Determine the geometric center point of each grid in the grid plane; Determining the centroid point of each grid according to the rasterized point cloud data; The grid feature information is determined based on the coordinate information of the geometric center point, the centroid point, and the detection points in each grid.

7. The method according to claim 6, characterized in that The coordinate information includes three-dimensional coordinate information corresponding to the detection point in the rasterized point cloud data, and two-dimensional coordinate information corresponding to the detection point in the grid plane; the grid feature information is determined based on the coordinate information of the geometric center point, the centroid point, and the detection points in each grid, including: Determine first difference information according to the coordinate information of the centroid and the three-dimensional coordinate information; Determine second difference information according to the coordinate information of the geometric center point and the two-dimensional coordinate information; The grid feature information is determined based on the first difference information, the second difference information and four-dimensional information of each of the detection points; wherein the four-dimensional information includes the three-dimensional coordinate information and laser reflection intensity information of the detection point in the point cloud data.

8. An electronic device, characterized in that: The electronic device comprises a determination unit and a detection unit; The determination unit is used to perform rasterization processing on the point cloud data corresponding to the detection area to obtain rasterized point cloud data; and determine a raster feature map based on the rasterized point cloud data, and determine a raster classification map and a height regression map according to the raster feature map; wherein the raster classification map is used to determine whether each raster is a ground raster; and the height regression map is used to determine the height information of each raster from the ground; The detection unit is used to perform ground detection processing based on the grid classification map and the height regression map to obtain a ground detection result corresponding to the detection area.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory storing instructions executable by the processor; when the instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

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

Images