A method and system for ground segmentation of a laser radar

Abstract

The application provides a laser radar ground segmentation method and system, wherein the method comprises the following steps: step 1, inputting laser radar point cloud data; step 2, pre-processing the laser radar point cloud data; step 3, constructing a data segmentation model, and inputting the pre-processed laser radar point cloud data into the data layer segmentation model; and step 4, outputting the segmented laser radar point cloud data. The application can effectively process the ground with a slope, improve the accuracy of ground point segmentation, and adaptively adjust the local slope threshold, so that the robustness and real-time performance are good, and the demand of automatic driving can be met.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of laser radar point cloud processing, and more particularly to a laser radar ground segmentation method and system. BACKGROUND

[0002] Laser radar has become the main sensor of the current automatic driving environment perception system due to its large data volume, high measurement accuracy, wide detection range, strong environmental adaptability and other characteristics. The point cloud data obtained by laser radar scanning not only includes obstacle point cloud, but also a large amount of ground point cloud. The ground point is often considered as redundant information for subsequent target detection, recognition and tracking tasks, so it is necessary to segment the ground point first.

[0003] The segmentation algorithm in the prior art has the risk of under-segmentation when the obstacle boundary point cloud coincides with the grid edge, and is easily affected by suspended obstacles. The global fitting method has poor real-time performance due to the calculation of multiple plane model parameters. The deep learning-based method has a large amount of labeled point cloud data and computing resources, and poor network interpretability, which limits the deployment of deep learning methods to real vehicles. SUMMARY

[0004] The purpose of the present application is to provide a laser radar ground segmentation method and system, which aims to solve the technical problems of the prior art that the ground segmentation algorithm has single specific scene, strong dependence on parameter setting, complex calculation and poor real-time performance.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is to provide a laser radar ground segmentation method, comprising:

[0006] Step 1: input laser radar point cloud data;

[0007] Step 2: pre-process the laser radar point cloud data;

[0008] Step 3: construct a data segmentation model, and input the pre-processed laser radar point cloud data into the data layer segmentation model;

[0009] Step 4: output the segmented laser radar point cloud data.

[0010] Preferably, step 3 comprises:

[0011] Step 3.1. The data segmentation model comprises a laser radar point cloud data coarse segmentation module and a laser radar data fine segmentation module;

[0012] Step 3.2. The pre-processed laser radar point cloud data is input into the laser radar point cloud data coarse segmentation module to obtain initial ground point cloud and initial non-ground point cloud;

[0013] Step 3.3. Input the initial ground point cloud and the initial non-ground point cloud obtained by rough segmentation into the laser radar point cloud data fine segmentation module to obtain the ground point cloud and the non-ground point cloud.

[0014] Preferably, step 3.2 comprises:

[0015] Step 3.2.1. Dividing the preprocessed laser radar point cloud data into a plurality of fan-shaped regions;

[0016] Step 3.2.2. Obtaining a local angle threshold of the preprocessed laser radar point cloud based on the divided fan-shaped regions;

[0017] Step 3.2.3. Obtaining the initial ground point cloud and the initial non-ground point cloud based on the local angle threshold.

[0018] Preferably, step 3.2.1 comprises:

[0019] Calculating the distance from the laser radar point cloud in the fan-shaped region to the origin of the laser radar point cloud spatial coordinate system based on the divided fan-shaped region;

[0020] The formula of the distance from the laser radar point cloud in the fan-shaped region to the origin of the laser radar point cloud spatial coordinate system is:

[0021]

[0022] Wherein, d i is the distance from the i-th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system, x i is the horizontal coordinate of the i-th laser radar point cloud, y i is the vertical coordinate of the i-th laser radar point cloud;

[0023] Based on the distance from the laser radar point cloud in the fan-shaped region to the origin of the laser radar point cloud coordinate system and the spatial coordinate system of the laser radar point cloud, the included angle of the adjacent laser radar point cloud and the vertical angle of the fan-shaped region origin and the slope surface are obtained; wherein, the calculation formula of the angle of the adjacent laser radar point cloud is:

[0024]

[0025] Wherein, ε i+1 is the included angle of the adjacent laser radar point cloud, Z i +1 is the vertical coordinate of the i+1-th laser radar point cloud, z i is the vertical coordinate of the i-th laser radar point cloud, d i +1 is the distance from the i+1-th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system, d i is the distance from the i-th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system;

[0026] The local angle threshold is obtained based on the included angle of adjacent laser radar point clouds.

[0027] The calculation formula of the local angle threshold is:

[0028]

[0029] is the local angle threshold of the mth sector, m is the sector of the laser radar point cloud, and ε is the angle of adjacent laser radar point clouds.

[0030] Preferably, step 3.2.3 comprises:

[0031] The laser radar point cloud judgment formula is obtained based on the local angle threshold, the type of the laser radar point cloud is judged based on the laser radar point cloud judgment formula, if the laser radar point cloud judgment formula is satisfied, it is a ground point cloud, and if the laser radar point cloud formula is not satisfied, it is a non-ground point cloud;

[0032] The laser radar point cloud judgment formula is:

[0033]

[0034] wherein θ g is the local angle threshold, z i is the vertical coordinate of the ith laser radar point cloud, x i is the horizontal coordinate of the ith laser radar point cloud, y i is the longitudinal coordinate of the ith laser radar point cloud, and H is the erection height of the laser radar.

[0035] Preferably, step 3.3 comprises:

[0036] Step 3.3.1. Project the initial ground point cloud and the initial non-ground point cloud into the RGB image;

[0037] Step 3.3.2. Calculate the vertical deflection angle and the horizontal deflection angle of the laser radar point cloud, and obtain the position and pixel value of the laser radar point cloud in the RGB image based on the vertical deflection angle and the horizontal deflection angle of the laser radar point cloud;

[0038] Step 3.3.3. Calculate the Euclidean distance of laser radar point clouds of different classifications, and obtain the normalized weight of the laser radar pixel point based on the Euclidean distance of laser radar point clouds of different classifications;

[0039] Step 3.3.4. Convolution filtering is performed on the RGB image based on the normalized weight of the laser radar pixel point;

[0040] Step 3.3.5. Project the coarsely segmented laser radar point cloud into the convolution filtered RGB image to obtain the ground point cloud and the non-ground point cloud.

[0041] The formula for the position of a LiDAR point cloud in an RGB image is:

[0042] R = (c, y)

[0043]

[0044]

[0045] Where R is the position of the LiDAR point cloud in the RGB image, c is the x-coordinate of the LiDAR point cloud in the RGB image, r is the y-coordinate of the LiDAR point cloud in the RGB image, and x... i Let y be the x-coordinate of the i-th lidar point cloud in the lidar spatial coordinate system. i Let z be the ordinate of the i-th lidar point cloud in the lidar spatial coordinate system. i Let ρ be the vertical coordinate of the i-th lidar point cloud in the lidar spatial coordinate system, and Δρ be the horizontal angular resolution of the lidar.

[0046] Preferably, step 3.3.3 includes:

[0047] A preset distance threshold is obtained based on the Euclidean distance of different categories of lidar point clouds, and the normalized weight of lidar pixels is obtained based on the preset distance threshold and the Euclidean distance of different categories of lidar point clouds.

[0048] The formula for the normalized weights of LiDAR pixels is:

[0049]

[0050]

[0051] Among them, w (r+i,c+i) W represents the unnormalized weights of the LiDAR point cloud pixels. (r+i,c+i) Let be the normalized weights of the LiDAR point cloud pixels, Δd be the Euclidean distance between LiDAR point clouds of different classifications, σ be the local effective range of the Gaussian kernel function, and Th be the weights of the normalized pixels. d This is a preset distance threshold.

[0052] Preferably, step 3.3.4 includes:

[0053] The formula for convolution filtering is:

[0054]

[0055]

[0056] Among them, Score r(r,c) represents the convolution result of the non-ground point cloud obtained from the coarse segmentation, and the score is... g (r,c) represents the convolution result of the ground point cloud obtained from coarse segmentation, W (r+i,c+i) is the normalization weight of the LiDAR point cloud pixels, c is the x-coordinate of the LiDAR point cloud in the RGB image, and r is the y-coordinate of the LiDAR point cloud in the RGB image.

[0057] If Score r (r,c)>Score g (r,c), then the corresponding lidar point cloud is a non-ground point, if Score r (r,c)≤Score g If (r,c), then the corresponding lidar point cloud is the ground point.

[0058] A lidar ground segmentation system, comprising:

[0059] The lidar point cloud acquisition module is used to acquire lidar point cloud data.

[0060] The preprocessing module is used to preprocess the lidar point cloud data;

[0061] The data segmentation model module is used to segment the preprocessed lidar point cloud data.

[0062] The lidar point cloud output module is used to acquire segmented lidar point cloud data.

[0063] The beneficial effects of the lidar ground segmentation method and system provided by this invention are as follows: Compared with the prior art, this invention can effectively handle sloping ground, improve the accuracy of ground point segmentation, and the local slope threshold can be adaptively adjusted, exhibiting strong robustness and good real-time performance, which can meet the needs of autonomous driving. Furthermore, this invention maintains high precision and recall in both simulation scenarios, and the average data consumption time per frame does not exceed 40ms, meeting the real-time requirements of autonomous driving. At the same time, the local slope threshold can be adaptively adjusted, exhibiting good robustness while maintaining good real-time performance, thus meeting the needs of practical engineering applications. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1A structural framework diagram of a laser radar ground segmentation method provided by an embodiment of the present application is provided.

[0066] Figure 2 A flowchart of a laser radar ground segmentation system provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0067] In order to make the technical problems to be solved by the present application, technical solutions and beneficial effects clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0068] Please refer to Figure 1 , now a laser radar ground segmentation method provided by the present application will be described.

[0069] A laser radar ground segmentation method, comprising the following steps:

[0070] Step 1: input laser radar point cloud data;

[0071] Step 2: pre-process the laser radar point cloud data;

[0072] Step 3: construct a data segmentation model, and input the pre-processed laser radar point cloud data into the data layer segmentation model;

[0073] Further, step 3 comprises:

[0074] Step 3.1. The data segmentation model comprises a laser radar point cloud data coarse segmentation module and a laser radar data fine segmentation module;

[0075] Step 3.2. The pre-processed laser radar point cloud data is input into the laser radar point cloud data coarse segmentation module to obtain initial ground point cloud and initial non-ground point cloud;

[0076] 360° is divided into 360 / Δρ fan-shaped regions according to the laser radar horizontal angle resolution Δρ, for example, the Rave C32-151A laser radar used by the present application has a horizontal angle resolution of 0.2°, and 1800 fan-shaped regions can be divided.

[0077] Step 3.2.1. divide the pre-processed laser radar point cloud data into fan-shaped regions;

[0078] Further, step 3.2.1 comprises:

[0079] Based on the pre-processed laser radar point cloud data, the laser radar horizontal angle resolution is obtained;

[0080] constructing a laser radar point cloud space coordinate system based on the preprocessed laser radar point cloud data;

[0081] calculating an included angle between the laser radar point cloud and the X-axis of the laser radar point cloud space coordinate system, and obtaining a sector region to which the laser radar point cloud belongs based on the included angle between the laser radar point cloud and the X-axis of the laser radar point cloud space coordinate system and a horizontal angle resolution of the laser radar;

[0082] The formula of the sector region is:

[0083]

[0084] wherein, S m is the sector region, Δρ is the horizontal angle resolution of the laser radar, x i is the horizontal coordinate of the laser radar point cloud, y i is the vertical coordinate of the laser radar point cloud.

[0085] Step 3.2.2. Obtaining a local angle threshold of the preprocessed laser radar point cloud based on the divided sector region;

[0086] Further, step 3.2.2 includes:

[0087] calculating a distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud space coordinate system based on the divided sector region;

[0088] The formula of the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud space coordinate system is:

[0089]

[0090] wherein, d i is the distance from the i-th laser radar point cloud to the origin of the laser radar point cloud space coordinate system, x i is the horizontal coordinate of the i-th laser radar point cloud, y i is the vertical coordinate of the i-th laser radar point cloud.

[0091] obtaining an included angle of adjacent laser radar point clouds and a vertical angle between the origin of the sector region and the slope surface based on the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud coordinate system and the space coordinate system of the laser radar; wherein, the calculation formula of the angle of the adjacent laser radar point clouds is:

[0092]

[0093] wherein, ε i+1 is the included angle of the adjacent laser radar point clouds, Z i +1 is the vertical coordinate of the i+1-th laser radar point cloud, z i is the vertical coordinate of the i-th laser radar point cloud, di +1 is the distance from the i+1th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system, d i is the distance from the i-th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system;

[0094] The local angle threshold is obtained based on the included angle of adjacent laser radar point clouds.

[0095] The calculation formula of the local angle threshold is:

[0096]

[0097] is the local angle threshold of the m-th sector region, m is the laser radar point cloud sector region, and ε is the angle of adjacent laser radar point clouds.

[0098] Step 3.2.3 obtains the ground point cloud and the non-ground point cloud based on the local angle threshold.

[0099] Further, step 3.2.3 includes:

[0100] The laser radar point cloud judgment formula is obtained based on the local angle threshold, the type of the laser radar point cloud is judged based on the laser radar point cloud judgment formula, if the laser radar point cloud judgment formula is satisfied, it is a ground point cloud, and if the laser radar point cloud formula is not satisfied, it is a non-ground point cloud;

[0101] The laser radar point cloud judgment formula is:

[0102]

[0103] wherein θ g is the local angle threshold, z i is the vertical coordinate of the i-th laser radar point cloud, x i is the horizontal coordinate of the i-th laser radar point cloud, y i is the longitudinal coordinate of the i-th laser radar point cloud, and H is the laser radar erection height.

[0104] Step 3.3. inputs the laser radar point cloud data after rough segmentation into a laser radar point cloud data fine segmentation module to obtain the segmented laser radar point cloud data.

[0105] Further, step 3.3 includes:

[0106] Step 3.3. inputs the initial ground point cloud and the initial non-ground point cloud obtained by rough segmentation into a laser radar point cloud data fine segmentation module to obtain the ground point cloud and the non-ground point cloud.

[0107] The RGB image size is R*C, R is the number of laser radar beams, C is the number of laser points generated by one laser radar emitter rotating 360°, the non-ground points marked by coarse segmentation are high-confidence obstacle points, the pixel value in the RGB image is red [255, 0, 0], and the marked ground points are high-confidence ground points, the pixel value in the RGB image is green [0, 255, 0], and the empty points are black.

[0108] Step 3.3.1. Project the initial ground point cloud and the initial non-ground point cloud into the RGB image;

[0109] Step 3.3.2. Obtain the position and pixel value of the laser radar point cloud in the RGB image based on the vertical deflection angle and the horizontal deflection angle of the laser radar point cloud;

[0110] The ground point cloud obtained by the first stage coarse segmentation contains part of the non-ground obstacle points, that is, over-segmentation occurs, and inaccurate segmentation usually occurs at the boundary between the ground point cloud and the obstacle point cloud. Therefore, the classification of the points near the red channel (C r ) representing the "high-confidence obstacle point cloud" and the green channel (C g ) representing the "high-confidence ground point cloud" needs to be re-identified, and this part of the point cloud is recorded as "unknown classification points".

[0111] Step 3.3.3. Calculate the Euclidean distance of the laser radar point cloud of different classifications, and obtain the normalized weight of the laser radar pixel point based on the Euclidean distance of the laser radar point cloud of different classifications;

[0112] Further, step 3.3.3 includes:

[0113] Based on the Euclidean distance of the laser radar point cloud of different classifications, a preset distance threshold is obtained, and based on the preset distance threshold and the Euclidean distance of the laser radar point cloud of different classifications, the normalized weight of the laser radar pixel point is obtained;

[0114] The formula of the normalized weight of the laser radar pixel point is:

[0115]

[0116]

[0117] where w (r+i,c+i) is the non-normalized weight of the laser radar point cloud pixel point, W (r+i,c+i) is the normalized weight of the laser radar point cloud pixel point, Δd is the Euclidean distance between the laser radar point clouds of different classifications, σ is the local action range of the Gaussian kernel function, and Thd is the preset distance threshold.

[0118] Step 3.3.4. Convolution filtering of the RGB image based on the normalized weight of the laser radar pixel point;

[0119] Further, step 3.3.4 includes:

[0120] The convolution filtering formula is:

[0121]

[0122]

[0123] Wherein, Score r (r,c) is the convolution result of the non-ground point cloud obtained by rough segmentation, Score g (r,c) is the convolution result of the ground point cloud obtained by rough segmentation, W (r+i,c+i) is the normalized weight of the laser radar point cloud pixel point, c is the horizontal coordinate of the laser radar point cloud in the RGB image, and r is the vertical coordinate of the laser radar point cloud in the RGB image.

[0124] If Score r (r,c) > Score g (r,c), the corresponding laser radar point cloud is a non-ground point cloud, and if Score r (r,c) ≤ Score g (r,c), the corresponding laser radar point cloud is a ground point cloud.

[0125] Step 3.3.5. Project the rough segmented laser radar point cloud into the convolution filtered RGB image to obtain the laser radar ground segmentation result.

[0126] The position formula of the laser radar point cloud in the RGB image is:

[0127] R = (c, y)

[0128]

[0129]

[0130] Wherein, R is the position of the laser radar point cloud in the RGB image, c is the horizontal coordinate of the laser radar point cloud in the RGB image, r is the vertical coordinate of the laser radar point cloud in the RGB image, x i is the horizontal coordinate of the i-th laser radar point cloud in the laser radar space coordinate system, y i is the vertical coordinate of the i-th laser radar point cloud in the laser radar space coordinate system, z i is the vertical coordinate of the i-th laser radar point cloud in the laser radar space coordinate system, and Δρ is the horizontal angle resolution of the laser radar.

[0131] The present application adopts simulation test to test the effectiveness and reliability of the segmentation result, artificially marks the data as non-ground points and ground points, refers to the confusion matrix, the precision P precision and recall P recall as evaluation indexes for verification, and the calculation formula is as follows:

[0132]

[0133]

[0134] Among them, TP represents the number of correct segmentation as ground points, FP represents the number of obstacle points missegmented as ground points, and FN represents the number of ground points missegmented as obstacle points. The precision P precision represents the proportion of true ground points in the point cloud segmented as ground points, and P recall Recall represents the proportion of true ground points in the calibration data extracted as ground points.

[0135] From the simulation data, 10 frames of data of flat road surface and slope road surface scene are selected to evaluate the segmentation effect and average time consumption of the algorithm. The results are shown in Table 1. Since the simulation data is more ideal than the measured data, the point cloud imaging quality is better, the precision and recall of the algorithm in the present application are maintained at a high value in the two simulation scenes, and the average time consumption of each frame of data is not more than 40 ms, which meets the real-time requirement of automatic driving.

[0136]

[0137] Table 1

[0138] The above is the introduction of the laser radar signal denoising method provided by the present application. In order to better implement the laser radar ground segmentation method provided by the present application, the present application also provides a laser radar ground segmentation system from the functional module angle.

[0139] According to Figure 2 , a laser radar ground segmentation system comprises:

[0140] A laser radar point cloud acquisition module is configured to acquire laser radar point cloud data.

[0141] A preprocessing module is configured to preprocess the laser radar point cloud data.

[0142] A data segmentation model module is configured to segment the preprocessed laser radar point cloud data.

[0143] A laser radar point cloud output module is configured to acquire the segmented laser radar point cloud data.

[0144] Compared with the prior art, the laser radar ground segmentation system has the same beneficial effects as the laser radar ground segmentation method, and details are not repeated here.

[0145] The above only describes preferred embodiments of the present application and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for ground segmentation of a laser radar, characterized by, The method comprises the following steps: Step 1: inputting laser radar point cloud data; Step 2: preprocessing the laser radar point cloud data; Step 3: constructing a data segmentation model, and inputting the preprocessed laser radar point cloud data into the data segmentation model; The step 3 comprises: Step 3.

1. the data segmentation model comprises a laser radar point cloud data coarse segmentation module and a laser radar data fine segmentation module; Step 3.

2. inputting the preprocessed laser radar point cloud data into the laser radar point cloud data coarse segmentation module to obtain initial ground point cloud and initial non-ground point cloud; The step 3.2 comprises: Step 3.2.

1. dividing the preprocessed laser radar point cloud data into sector regions; The step 3.2.1 comprises: obtaining a laser radar horizontal angle resolution based on the preprocessed laser radar point cloud data; constructing a laser radar point cloud space coordinate system based on the preprocessed laser radar point cloud data; calculating an included angle between the laser radar point cloud and an X-axis of the laser radar point cloud space coordinate system, and obtaining a sector region to which the laser radar point cloud belongs based on the included angle between the laser radar point cloud and the X-axis of the laser radar point cloud space coordinate system and the laser radar horizontal angle resolution; a formula of the sector region is: ； wherein S m is the fan-shaped region, is the horizontal angle resolution of the lidar, x i is the horizontal coordinate of the lidar point cloud, y i is the vertical coordinate of the lidar point cloud; Step 3.2.

2. obtaining a local angle threshold of the preprocessed laser radar point cloud based on the divided sector regions; The step 3.2.2 comprises: calculating a distance from the laser radar point cloud in the sector region to an origin of the laser radar point cloud space coordinate system based on the divided sector regions; obtaining an included angle of adjacent laser radar point clouds and a vertical angle between a sector region origin and a slope surface based on the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud coordinate system and a space coordinate system of the laser radar point cloud; obtaining the local angle threshold based on the included angle of the adjacent laser radar point clouds; a formula of the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud space coordinate system is: ； in, d i For the first i The distance from each lidar point cloud to the origin of the lidar point cloud spatial coordinate system x i For the first i The x-coordinate of a lidar point cloud. y i For the first i The vertical coordinate of a lidar point cloud; a calculation formula of the included angle of the adjacent laser radar point clouds is: ； wherein, is an angle of adjacent laser radar point clouds, z i+1 is a vertical coordinate of the first i +1thlaser radar point cloud, z i is a vertical coordinate of the first i +1thlaser radar point cloud, d i+1 is a distance from the first i +1thlaser radar point cloud to the origin of the laser radar point cloud spatial coordinate system, d i is a distance from the first i +1thlaser radar point cloud to the origin of the laser radar point cloud spatial coordinate system; a calculation formula of the local angle threshold is: ； is a local angle threshold for the mth sector region, m being the sector region of the laser radar point cloud, is an angle of an adjacent laser radar point cloud; Step 3.2.

3. obtaining the initial ground point cloud and the initial non-ground point cloud based on the local angle threshold; The step 3.2.3 comprises: obtaining a laser radar point cloud judgment formula based on the local angle threshold, judging a type of the laser radar point cloud based on the laser radar point cloud judgment formula, if the laser radar point cloud judgment formula is satisfied, the initial ground point cloud is obtained, and if the laser radar point cloud formula is not satisfied, the initial non-ground point cloud is obtained; the laser radar point cloud judgment formula is: ； in, This is a local angle threshold. z i For the first i The vertical coordinates of a lidar point cloud. x i For the first i The x-coordinate of a lidar point cloud. y i For the first i The ordinate of a LiDAR point cloud. H The height for setting up the lidar; Step 3.

3. inputting the initial ground point cloud and the initial non-ground point cloud obtained through the coarse segmentation into the laser radar point cloud data fine segmentation module to obtain ground point cloud and non-ground point cloud; The step 3.3 comprises: Step 3.3.

1. projecting the initial ground point cloud and the initial non-ground point cloud into an RGB image; Step 3.3.

2. calculating a vertical deflection angle and a horizontal deflection angle of the laser radar point cloud, and obtaining a position and a pixel value of the laser radar point cloud in the RGB image based on the vertical deflection angle and the horizontal deflection angle of the laser radar point cloud; Step 3.3.

3. Calculate the Euclidean distance of the laser radar point cloud of different classifications, and obtain the normalized weight of the laser radar pixel point based on the Euclidean distance of the laser radar point cloud of different classifications; The step 3.3.3 comprises: Obtain a preset distance threshold based on the Euclidean distance of the laser radar point cloud of different classifications, and obtain the normalized weight of the laser radar pixel point based on the preset distance threshold and the Euclidean distance of the laser radar point cloud of different classifications; The formula of the normalized weight of the laser radar pixel point is: ； wherein, is an unnormalized weight of a laser radar point cloud pixel point, is a normalized weight of a laser radar point cloud pixel point, is an Euclidean distance between laser radar point clouds of different classifications, is a local action range of a Gaussian kernel function, Th d is a preset distance threshold. Step 3.3.

4. Convolution filtering is performed on the RGB image based on the normalized weight of the laser radar pixel point; Step 3.3.

5. Project the laser radar point cloud after the coarse segmentation into the RGB image after the convolution filtering, and obtain the laser radar ground segmentation result; The formula of the position of the laser radar point cloud in the RGB image is: ； in, R This indicates the location of the LiDAR point cloud in the RGB image. c Let x be the x-coordinate of the LiDAR point cloud in the RGB image. r Let y be the vertical coordinate of the LiDAR point cloud in the RGB image. x i For the first i The x-coordinate of a point cloud from a lidar radar in the lidar spatial coordinate system. y i For the first i The ordinate of a point cloud from a lidar radar in the lidar spatial coordinate system. z i For the first i The vertical coordinates of a point cloud from a lidar system in the lidar spatial coordinate system. This refers to the horizontal angular resolution of the lidar. Step 4: Output the segmented laser radar point cloud data.

2. The method of claim 1, wherein, The step 3.3.4 comprises: The formula of the convolution filtering is: ； wherein, is a convolution result of the non-ground point cloud obtained by coarse segmentation, is a convolution result of the ground point cloud obtained by coarse segmentation, is a normalized weight of a pixel point of the lidar point cloud, c is a horizontal coordinate of the lidar point cloud in the RGB image, r is a vertical coordinate of the lidar point cloud in the RGB image. If , the corresponding lidar point cloud is a non-ground point cloud, and if , the corresponding lidar point cloud is a ground point cloud.

3. A lidar ground segmentation system, comprising: Comprise: The laser radar point cloud acquisition module is used for acquiring the laser radar point cloud data; The preprocessing module is used for preprocessing the laser radar point cloud data; The data segmentation model module is used for data segmentation on the preprocessed laser radar point cloud data; Wherein, the data segmentation on the preprocessed laser radar point cloud data comprises: Step 3.

1. The data segmentation model comprises a laser radar point cloud data coarse segmentation module and a laser radar data fine segmentation module; Step 3.

2. The preprocessed laser radar point cloud data is input into the laser radar point cloud data coarse segmentation module to obtain initial ground point cloud and initial non-ground point cloud; The step 3.2 comprises: Step 3.2.

1. Divide the preprocessed laser radar point cloud data into sector regions; The step 3.2.1 comprises: Obtain the laser radar horizontal angle resolution based on the preprocessed laser radar point cloud data; Construct a laser radar point cloud space coordinate system based on the preprocessed laser radar point cloud data; Calculate the included angle between the laser radar point cloud and the X axis of the laser radar point cloud space coordinate system, and obtain the sector region to which the laser radar point cloud belongs based on the included angle between the laser radar point cloud and the X axis of the laser radar point cloud space coordinate system and the laser radar horizontal angle resolution; The formula of the sector region is: ； wherein S m is the fan-shaped region, is the horizontal angle resolution of the lidar, x i is the horizontal coordinate of the lidar point cloud, y i is the vertical coordinate of the lidar point cloud; Step 3.2.

2. Obtain the local angle threshold of the preprocessed laser radar point cloud based on the divided sector region; The step 3.2.2 comprises: Calculate the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud space coordinate system based on the divided sector region; Obtain the included angle of adjacent laser radar point clouds and the vertical angle between the sector region origin and the slope surface based on the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud coordinate system and the space coordinate system of the laser radar point cloud; Obtain the local angle threshold based on the included angle of the adjacent laser radar point clouds; The formula of the distance from the laser radar point cloud in the sector region to the origin of the laser radar point cloud space coordinate system is: ； in, d i For the first i The distance from each lidar point cloud to the origin of the lidar point cloud spatial coordinate system x i For the first i The x-coordinate of a lidar point cloud. y i For the first i The vertical coordinate of a lidar point cloud; The calculation formula of the included angle of the adjacent laser radar point clouds is: ； wherein, is an angle of adjacent laser radar point clouds, z i+1 is a vertical coordinate of the first i +1th laser radar point cloud, z i is a vertical coordinate of the first i +1th laser radar point cloud, d i+1 is a distance from the first i +1th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system, d i is a distance from the first i +1th laser radar point cloud to the origin of the laser radar point cloud spatial coordinate system; The calculation formula of the local angle threshold is: ； is a local angle threshold for the mth sector region, m being the sector region of the laser radar point cloud, is an angle of an adjacent laser radar point cloud; Step 3.2.

3. obtaining an initial ground point cloud and an initial non-ground point cloud based on the local angle threshold; The step 3.2.

3. comprises: obtaining a laser radar point cloud judgment formula based on the local angle threshold, judging the type of the laser radar point cloud based on the laser radar point cloud judgment formula, if the laser radar point cloud judgment formula is met, it is an initial ground point cloud, if the laser radar point cloud formula is not met, it is an initial non-ground point cloud; The laser radar point cloud judgment formula is: ； wherein, is a local angular threshold, z i is a vertical coordinate of the i-th laser radar point cloud, i x i is a horizontal coordinate of the i-th laser radar point cloud, i y i is a longitudinal coordinate of the i-th laser radar point cloud, i H is a laser radar mounting height;​​​ Step 3.

3. inputting the initial ground point cloud and the initial non-ground point cloud obtained by the coarse segmentation into a laser radar point cloud data fine segmentation module to obtain a ground point cloud and a non-ground point cloud; The step 3.

3. comprises: Step 3.3.

1. projecting the initial ground point cloud and the initial non-ground point cloud into an RGB image; Step 3.3.

2. calculating a vertical deflection angle and a horizontal deflection angle of the laser radar point cloud, and obtaining a position and a pixel value of the laser radar point cloud in the RGB image based on the vertical deflection angle and the horizontal deflection angle of the laser radar point cloud; Step 3.3.

3. calculating the Euclidean distance of the laser radar point cloud of different classifications, and obtaining the normalized weight of the laser radar pixel point based on the Euclidean distance of the laser radar point cloud of different classifications; The step 3.3.

3. comprises: obtaining a preset distance threshold based on the Euclidean distance of the laser radar point cloud of different classifications, and obtaining the normalized weight of the laser radar pixel point based on the preset distance threshold and the Euclidean distance of the laser radar point cloud of different classifications; The formula of the normalized weight of the laser radar pixel point is: ； wherein, is an unnormalized weight of a laser radar point cloud pixel point, is a normalized weight of a laser radar point cloud pixel point, is an Euclidean distance between laser radar point clouds of different classifications, is a local action range of a Gaussian kernel function, Th d is a preset distance threshold. Step 3.3.

4. performing convolution filtering on the RGB image based on the normalized weight of the laser radar pixel point; Step 3.3.

5. projecting the laser radar point cloud after the coarse segmentation into the RGB image after the convolution filtering to obtain a laser radar ground segmentation result; The formula of the position of the laser radar point cloud in the RGB image is: ； in, R This indicates the location of the LiDAR point cloud in the RGB image. c Let x be the x-coordinate of the LiDAR point cloud in the RGB image. r Let y be the vertical coordinate of the LiDAR point cloud in the RGB image. x i For the first i The x-coordinate of a point cloud from a lidar radar in the lidar spatial coordinate system. y i For the first i The ordinate of a point cloud from a lidar radar in the lidar spatial coordinate system. z i For the first i The vertical coordinates of a point cloud from a lidar system in the lidar spatial coordinate system. This refers to the horizontal angular resolution of the lidar. A laser radar point cloud output module is configured to obtain the segmented laser radar point cloud data.

Images