Ground segmentation method and system based on adaptive grid division and intensity information

Abstract

The invention belongs to the technical field of point cloud segmentation. The invention provides a ground segmentation method and system based on adaptive grid division and intensity information, and the method comprises the steps: carrying out the height and intensity preprocessing of laser radar point cloud data, removing noise and invalid point cloud, and obtaining the preprocessed point cloud data; carrying out adaptive grid division on the preprocessed point cloud data to obtain plane grids; performing grid area growth on the plane grid to obtain a ground grid; dynamically grouping the ground grids based on the grids of the intensity information to obtain intensity grid categories; and carrying out confidence coefficient correction on the strength grid category to obtain a ground point and a non-ground point. The method can effectively identify and segment ground conditions such as roads and slopes, eliminates interference point clouds such as roadside grassland, and improves the accuracy of ground segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud segmentation, and in particular to a ground segmentation method and system based on adaptive grid division and intensity information. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, the application of mobile robots has expanded, and their operational scenarios have become increasingly complex. SLAM (Simultaneous Localization and Mapping) is currently a hot topic for solving complex indoor and outdoor environmental perception problems for mobile robots. Ground segmentation of LiDAR point clouds is a fundamental task in environmental perception. Its primary goal is to distinguish ground points from non-ground points, and its accuracy directly impacts positioning and mapping.

[0004] Currently, common methods for ground segmentation of laser point clouds are mainly based on point feature analysis and fixed grid division. The point feature analysis method determines whether each point is a ground point by calculating its normal vector, height information, curvature, etc.; the fixed grid division method commonly uses fixed grids of different sizes based on the distance from the laser radar and based on the number of point clouds. The point cloud in the grid is then analyzed to estimate the ground points.

[0005] However, due to the complexity of the working scene, point feature analysis-based methods require long point-by-point calculations and are sensitive to noise points and sparse areas, which can easily lead to misjudgments. Fixed grid partitioning methods use fixed grid sizes that cannot adapt to the local distribution characteristics of point clouds, which can easily lead to missed detections. Furthermore, both of these methods can incorrectly segment point clouds with areas similar to the ground (such as roadside grass) as the ground, resulting in low segmentation accuracy. Summary of the Invention

[0006] In order to address the shortcomings of the existing technology, the present invention provides a ground segmentation method and system based on adaptive grid division and intensity information, which can effectively identify and segment ground conditions such as roads and slopes, while eliminating interfering point clouds such as roadside grass, thereby improving the accuracy of ground segmentation.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a ground segmentation method based on adaptive grid division and intensity information.

[0008] A ground segmentation method based on adaptive grid division and intensity information includes the following steps: Get lidar point cloud data; Performing height and intensity preprocessing on the laser radar point cloud data to remove noise and invalid point clouds to obtain preprocessed point cloud data; Adaptively divide the pre-processed point cloud data into plane grids; Performing grid region growing on the plane grid to obtain a ground grid; Dynamically grouping the ground grid based on intensity information to obtain intensity grid categories; The intensity grid categories are corrected for confidence to obtain ground points and non-ground points.

[0009] As a further limitation of the first aspect of the present invention, adaptively dividing the preprocessed point cloud data into a plane grid includes: Based on the pre-processed point cloud data, on the horizontal plane, with the lidar as the center, an initial grid of the same size is divided around the lidar; Calculate the number of point clouds in each initial grid and determine whether the number of point clouds N in each initial grid is greater than the set threshold. If not, delete the point clouds in the initial grid; if so, retain the point clouds in the initial grid. For the grid where the point cloud is retained, calculate the covariance matrix of the point cloud; Perform eigenvalue decomposition on the covariance matrix and obtain three eigenvalues and its corresponding eigenvector 、 and ; If the smallest eigenvalue Less than the eigenvalue and eigenvalues , and flatness Less than the flatness threshold , indicating that the point cloud is distributed on a plane, then the corresponding initial grid is not divided and is directly marked as a plane grid. Otherwise, the centroid of the calculated initial grid is used. As the center, it is divided into four sub-grids; When the number of divisions of a sub-grid is greater than the maximum number of divisions, or when the number of point clouds in a sub-grid is less than or equal to the set threshold, the sub-grid is marked as a non-ground grid, and the point cloud in the non-ground grid is a non-ground point; otherwise, for the grid whose point cloud is retained, continue to calculate the covariance matrix of the point cloud.

[0010] As a further limitation of the first aspect of the present invention, the covariance matrix P comprises: ,in, Indicates the first i The coordinates of the points, is the centroid of the point cloud within the grid, and N is the number of points in the grid.

[0011] As a further limitation of the first aspect of the present invention, flatness ,include: ,in, is the flatness threshold.

[0012] As a further limitation of the first aspect of the present invention, performing grid region growing on the plane grid to obtain a ground grid includes: Determine the initial seed grid, initialize the seed grid sequence, and add the initial seed grid to the seed area. Perform normal vector consistency and flatness constraint judgment on the neighboring grids of the seed grid in turn, and calculate the angle between the normal vector of the neighboring grid and the normal vector of the current seed grid. , if the normal vector angle Less than threshold , the direction of the neighboring grid is considered to be consistent with the ground direction, and this neighboring grid is added to the seed area. If the flatness of this neighboring grid is less than the set threshold, this neighboring grid is added to the seed grid sequence as the seed grid for subsequent expansion, the current seed grid is deleted, and a seed grid in the seed grid sequence is selected to continue the normal vector consistency and flatness constraint judgment until the seed grid sequence is empty; After the region growing is completed, the normal vectors of all the obtained seed regions are recalculated, and the angle between the normal vector and the set ground normal vector is solved. If the angle is greater than the set threshold, the grid in the seed region is marked as a non-ground point grid and removed from the plane grid. The remaining grids are marked as ground grids.

[0013] As a further limitation of the first aspect of the present invention, determining the initial seed grid includes: Select the grids within the set distance range in front, behind, left and right of the LiDAR from the plane grid as candidate grids for the initial seed grid, and calculate the average height of the point cloud in each candidate grid If the average height Greater than the height information threshold , it is determined that it is not a ground grid. After removing all ground grids, the flatness of the remaining candidate grids is calculated respectively, and they are arranged in ascending order of flatness. The grid with the smallest flatness is selected as the initial seed grid.

[0014] As a further limitation of the first aspect of the present invention, the ground grid is dynamically grouped based on the intensity information to obtain intensity grid categories, including: Traverse all ground grids and calculate the average intensity value of each ground grid; Sort the average intensity values ​​of all ground grids from small to large to obtain an ordered sequence: Traverse the ordered sequence of intensity values, starting from the first ground grid with the smallest average intensity value, and group together the ground grids whose average intensity difference between adjacent ground grids is less than the intensity difference threshold. If the intensity difference is greater than the intensity difference threshold, start a new grouping; All ground grids are grouped and each group is defined as an intensity grid class.

[0015] As a further limitation of the first aspect of the present invention, the average intensity value ,include: ,in, For the grid i The intensity value of each point, N is the total number of point clouds in the grid.

[0016] As a further limitation of the first aspect of the present invention, performing confidence correction on the intensity grid categories to obtain ground points and non-ground points includes: The intensity grid category containing the largest number of point clouds is the high-confidence intensity grid. The point clouds contained in the high-confidence intensity grid are all ground points, and the remaining intensity grids are low-confidence intensity grids. For any low-confidence grid, if there are more than three high-confidence grids in the neighborhood of this low-confidence grid, then this low-confidence grid will be corrected to a high-confidence grid, otherwise no correction will be performed; After confidence correction, the point clouds in the high-confidence grids are marked as ground point clouds, and the point clouds in the remaining grids are marked as non-ground point clouds.

[0017] In a second aspect, the present invention provides a ground segmentation system based on adaptive grid division and intensity information.

[0018] A ground segmentation system based on adaptive grid division and intensity information, comprising: The point cloud acquisition unit is configured to: acquire laser radar point cloud data; a point cloud preprocessing unit configured to: perform height and intensity preprocessing on the laser radar point cloud data, remove noise and invalid point clouds, and obtain preprocessed point cloud data; The grid division unit is configured to: perform adaptive grid division on the pre-processed point cloud data to obtain a plane grid; A grid region growing unit is configured to: perform grid region growing on the plane grid to obtain a ground grid; The grid dynamic grouping unit is configured to: dynamically group the ground grids based on the intensity information to obtain intensity grid categories; The grid confidence correction unit is configured to perform confidence correction on the intensity grid category to obtain ground points and non-ground points.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention innovatively proposes a ground segmentation strategy based on adaptive grid division and intensity information. It uses adaptive grid division to associate laser point clouds with flatness, effectively reducing the amount of calculation and improving computational efficiency. At the same time, it avoids the point cloud sparseness problem in some areas, thereby improving the rationality of the grid division of point cloud data.

[0020] 2. The present invention obtains a ground grid by performing region growing on the plane grid and then performing normal vector processing, effectively extracting ground planes such as road surfaces and slopes. At the same time, the ground grid is dynamically grouped based on intensity information to obtain an intensity grid, thereby accurately extracting ground points and effectively marking point clouds outside the ground that are easily mistakenly selected as ground points.

[0021] 3. The present invention corrects the confidence of the intensity grid, distinguishes ground points and non-ground points more accurately, and improves the accuracy of ground segmentation.

[0022] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 A schematic flow chart of a ground segmentation method based on adaptive grid division and intensity information provided in Example 1 of the present invention; Figure 2 A schematic diagram of the process of adaptive grid division provided in Example 1 of the present invention; Figure 3 This is a diagram showing the effect of adaptive grid division provided in Example 1 of the present invention; Figure 4 A schematic diagram of the process of grid region growth provided in Example 1 of the present invention; Figure 5 A schematic diagram of the process of dynamic grid grouping based on intensity information provided in Example 1 of the present invention; Figure 6 A schematic diagram of intensity grid confidence correction provided in Example 1 of the present invention; Figure 7 A schematic diagram of a ground segmentation system based on adaptive grid division and intensity information provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0027] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0028] Example 1: This implementation proposes a ground segmentation method based on adaptive grid division and intensity information, such as Figure 1 As shown, the process includes the following: obtaining lidar point cloud data; performing height and intensity preprocessing on the point cloud data to remove noise and invalid point clouds; adaptive grid division to obtain plane grids; grid region growth to obtain ground grids; dynamic grouping of grids based on intensity information to obtain intensity grids; and intensity grid confidence correction to obtain ground points and non-ground points.

[0029] For a given LiDAR point cloud data, some noise points are occasionally observed below the actual ground surface. These noise points will affect the subsequent process. Therefore, the main goal of the data preprocessing part is to filter out the noise. Since the noise points are usually below the actual ground surface and the intensity is small due to the large number of reflections, the noise height threshold and noise intensity threshold are used to remove the noise. The noise intensity threshold Initial setting is 0.2, the noise height threshold , The installation height of the LiDAR on the robot; Point cloud intensity is affected by the angle of incidence and distance. The smaller the angle of incidence, the greater the intensity, and the closer the distance, the greater the intensity. This can cause significant variations in the intensity of point clouds of the same material, adversely affecting subsequent grid division. Therefore, point clouds within 1 meter are removed. Furthermore, since the point cloud data scanned by the LiDAR is generally within 20 meters, point cloud data beyond 20 meters is removed. Through these processes, a valid point cloud collection is obtained.

[0030] In this implementation, the main goal of adaptive grid division is to divide the current point cloud data into grids. Through covariance analysis and eigenvalue decomposition of the point cloud, the grid division is dynamically adjusted, and the plane grid composed of plane points is selected from it, which facilitates subsequent processing on the plane points, reduces the number of processed point clouds, and improves computational efficiency.

[0031] Specifically, such as Figure 2 As shown in Figure 2, the specific process of adaptive gridding includes: Step S1: After preprocessing, the point cloud data is divided into the following areas on the xy plane: The initial grid.

[0032] Step S2: Calculate the number of point clouds in each initial grid and determine the number of point clouds in each grid Is it greater than 20 (i.e. set threshold), if , then continue; otherwise delete the point cloud in the grid. This step can effectively remove the interference of sparse areas and reduce the computational complexity.

[0033] Step S3: For the retained grid, further calculate the covariance matrix P of its point cloud to describe the distribution characteristics of the point cloud within the grid: (1); in, represents the coordinates of the i-th point in the grid, is the centroid of the point cloud within the grid, express The transpose of .

[0034] By performing eigenvalue decomposition on the covariance matrix, we get three eigenvalues: , and its corresponding eigenvector 、 、 .

[0035] Step S4: Determine whether the grid is a planar grid and whether further division is required. The division is based on the following flatness constraint: If the minimum eigenvalue Much smaller than the other two eigenvalues 、 , and flatness Less than the flatness threshold , indicating that the point cloud is distributed on a plane, the grid is not divided and it is marked as a plane grid. Otherwise, the centroid of the grid is calculated. The grid division based on the centroid position of the point cloud can cope with complex environments and improve segmentation accuracy.

[0036] (2); in, Set to 0.01.

[0037] Step S5: In order to control the amount of calculation, the maximum number of sub-grid divisions is greater than 4 times (i.e., the maximum number of divisions), or when the number of point clouds in a sub-grid is less than 20 (i.e., the set threshold), the sub-grid is marked as a non-ground grid, and the point cloud therein is a non-ground point; otherwise, the operations of steps S3 and S4 are repeated. After this adaptive grid division, grids with significant planar features can be effectively extracted, reducing the interference of non-ground grids on subsequent processing. The segmentation effect is shown in the figure below. Figure 3 shown.

[0038] After completing the adaptive grid division, a certain number of planar grids that meet the flatness requirements are obtained. However, these grids are not completely ground grids because they may contain some isolated flat areas (such as the top of a car or a tabletop) and non-ground grids with varying heights (such as steps). Therefore, a region growing algorithm is further used to expand the ground grid from the planar grid and filter out the non-ground grids.

[0039] The core idea of ​​region growing is to use the initial seed grid as the starting point, and gradually add grids similar to the seed grid to the ground area based on the characteristics of its neighboring grids. The implementation process of grid region growing is as follows: Figure 4 As shown, the following process is included: Step (1): Selection of initial seed grid.

[0040] The present invention first selects grids within a 4-meter radius of the LiDAR from a planar grid as candidate initial seed grids. The grids are then sorted in ascending order based on their flatness, and the grid with the lowest flatness is selected as the initial seed grid. However, relying solely on flatness can lead to misselection. For example, non-ground areas such as the top of a vehicle may have very low flatness. To avoid this, height information filtering is introduced.

[0041] Height information filtering: Calculate the average height of the point cloud in the grid , if it is greater than the height information threshold , then the grid is considered not to be a ground grid, that is: (3); in, Take 0.1.

[0042] After removing non-ground grids that do not meet the height information, the grid with smaller flatness is used as the initial seed grid.

[0043] Step (2): Grid region growing: After selecting the initial seed grid, the ground area is expanded through the region growing algorithm. The expansion process mainly determines whether to add the neighboring grid to the seed area based on its characteristics.

[0044] Initialize the seed grid sequence, add the initial seed grid to the seed area, and perform normal vector consistency and flatness constraint judgment on the neighboring grids of the seed grid in turn.

[0045] Normal vector consistency: Calculate the angle between the normal vector of the neighboring grid and the normal vector of the current seed grid , if the angle is less than the threshold , the neighboring grid direction is considered to be consistent with the ground direction and is added to the seed region: (4); in, is the current seed grid normal, is the neighborhood grid normal vector, Pick .

[0046] If, on this basis, the flatness of the neighborhood grid is less than the set threshold of 0.005, it will also be added to the seed grid sequence as a seed for subsequent expansion, and the current seed grid will be deleted. The above operation will be repeated until the seed grid sequence is empty.

[0047] Step (3): Normal vector correction: After the region growing is completed, recalculate the normal vectors of all the obtained seed regions and solve the normal vectors and the set ground normal vectors. If the angle is greater than 45°, the grids in the seed area are marked as non-ground point grids, removed from the plane grid, and the remaining grids are marked as ground grids.

[0048] The region growing algorithm can provide data support for subsequent processing based on intensity information.

[0049] After region growing, although the ground grid has been roughly extracted, some ground grids may still contain some misclassified non-ground points (such as vegetation or obstacles on the road). Therefore, the intensity information is further used to dynamically group ground grids with similar intensity values.

[0050] Dynamic grouping of grids based on intensity information Figure 5 Specifically, the process includes the following: Step (A): Intensity value calculation.

[0051] Iterate over all ground grids and calculate the average intensity value of each grid: (5); in, is the intensity value of the i-th point in the grid, and N is the total number of point clouds in the grid. Describes the overall intensity value of the grid.

[0052] Step (B): Sort in ascending order.

[0053] Sort the average intensity values ​​of all grids from small to large to get an ordered sequence : (6); Among them, i is the sorted index, is the average intensity value of the first grid, is the average intensity value of the second grid, is the average intensity value of the mth grid.

[0054] Step (C): Dynamic grouping.

[0055] Set an intensity difference threshold , indicating that the grids with adjacent grid intensity values ​​within the allowable range belong to the same group. Traverse the ordered sequence of intensity values, starting from the first grid, and group the grids with adjacent grid intensity values ​​less than The grids are gathered together, that is , if the strength difference is greater than , start a new group; among them, is the average intensity value of the i-th grid, is the average intensity value of the i+1th grid.

[0056] Step (D): Generate intensity categories.

[0057] Follow steps (A)-(C) to complete the grouping of all rasters, with each group defined as an intensity raster category.

[0058] Since the intensity values ​​of different materials vary significantly, non-ground points can be clearly distinguished by intensity values. However, in actual applications, due to the diversity of ground materials, such as patched ground and ground with icons, the intensity values ​​of some areas may vary greatly, leading to mis-segmentation. Therefore, confidence correction is introduced to the intensity grid to ensure accurate ground segmentation results. The specific process of confidence correction includes: Step (a): Define the high confidence grid.

[0059] The intensity grid containing the largest number of point clouds is defined as a high-confidence grid. The point clouds in this category are considered to be ground points, and the remaining intensity grids are defined as low-confidence grids.

[0060] Step (b): Low confidence grid correction.

[0061] For a low-confidence grid, if there are more than three high-confidence grids in its neighborhood, it will be corrected to a high-confidence grid, otherwise no correction will be performed.

[0062] Step (c): Generate the final ground point cloud.

[0063] After confidence correction, the point cloud in the high confidence grid is marked as ground point cloud, and the point cloud in the remaining grids is marked as non-ground point cloud. The correction is as follows: Figure 6 shown.

[0064] Example 2: like Figure 7 As shown, this implementation provides a ground segmentation system based on adaptive grid division and intensity information, including: The point cloud acquisition unit is configured to: acquire laser radar point cloud data; a point cloud preprocessing unit configured to: perform height and intensity preprocessing on the laser radar point cloud data, remove noise and invalid point clouds, and obtain preprocessed point cloud data; The grid division unit is configured to: perform adaptive grid division on the pre-processed point cloud data to obtain a plane grid; A grid region growing unit is configured to: perform grid region growing on the plane grid to obtain a ground grid; The grid dynamic grouping unit is configured to: dynamically group the ground grids based on the intensity information to obtain intensity grid categories; The grid confidence correction unit is configured to perform confidence correction on the intensity grid category to obtain ground points and non-ground points.

[0065] The specific working process of each of the above units is described in Example 1 and will not be repeated here.

[0066] It is understandable that each of the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (or some) of the units can be further divided into multiple functionally smaller units to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In actual applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In actual applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0067] According to another embodiment of the present application, the system described in this embodiment can be constructed and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0068] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A ground segmentation method based on adaptive grid division and intensity information, characterized in that: The process includes: Get LiDAR point cloud data; Performing height and intensity preprocessing on the laser radar point cloud data to remove noise and invalid point clouds, thereby obtaining preprocessed point cloud data; Adaptively grid the preprocessed point cloud data to obtain a plane grid; Performing grid region growth on the plane grid to obtain a ground grid; Dynamically grouping the ground grid based on the intensity information to obtain intensity grid categories; The intensity grid categories are confidence modified to obtain ground points and non-ground points.

2. The ground segmentation method based on adaptive grid division and intensity information according to claim 1, characterized in that: The pre-processed point cloud data is adaptively gridded to obtain a plane grid, including: According to the pre-processed point cloud data, on the horizontal plane, with the laser radar as the center, divide the initial grid of the same size around the laser radar; Calculate the number of point clouds in each initial grid, and determine whether the number of point clouds N in each initial grid is greater than a set threshold. If not, delete the point clouds in the initial grid; if yes, retain the point clouds in the initial grid. For the grid whose point cloud is retained, calculate the covariance matrix of the point cloud; Perform eigenvalue decomposition on the covariance matrix and obtain three eigenvalues and its corresponding eigenvector , and ; If the smallest eigenvalue Less than the eigenvalue and eigenvalues , and the flatness Less than flatness threshold , indicating that the point cloud is distributed on a plane, the corresponding initial grid is not divided and is directly marked as a plane grid. Otherwise, the centroid of the calculated initial grid is used. As the center, it is divided into four sub-grids; When the number of divisions of a sub-grid is greater than the maximum number of divisions, or when the number of point clouds in a sub-grid is less than or equal to the set threshold, the sub-grid is marked as a non-ground grid, and the point cloud in the non-ground grid is a non-ground point; otherwise, for the grid where the point cloud is retained, continue to calculate the covariance matrix of the point cloud.

3. The ground segmentation method based on adaptive grid division and intensity information as claimed in claim 2, characterized in that: The covariance matrix P, including: ,in, represents the coordinates of the i-th point in the grid, is the centroid of the point cloud in the grid, and N represents the number of points in the grid.

4. The ground segmentation method based on adaptive grid division and intensity information as claimed in claim 2, characterized in that: Flatness ,include: ,in, is the flatness threshold.

5. The ground segmentation method based on adaptive grid division and intensity information as claimed in claim 1, characterized in that: Performing grid region growth on the plane grid to obtain a ground grid includes: Determine the initial seed grid, initialize the seed grid sequence, and add the initial seed grid to the seed area. Perform normal vector consistency and flatness constraint judgment on the neighboring grids of the seed grid in turn, and calculate the angle between the normal vector of the neighboring grid and the normal vector of the current seed grid. , if the normal vector angle Less than threshold , the direction of the neighborhood grid is considered to be consistent with the ground direction, and the neighborhood grid is added to the seed area. If the flatness of the neighborhood grid is less than the set threshold, the neighborhood grid is added to the seed grid sequence as the seed grid for subsequent expansion, the current seed grid is deleted, and a seed grid in the seed grid sequence is selected to continue to execute the normal vector consistency and flatness constraint judgment until the seed grid sequence is empty; After the region growing is completed, the normal vectors of all the obtained seed regions are recalculated, and the angle between the normal vector and the set ground normal vector is solved. If the angle is greater than the set threshold, the grid in the seed region is marked as a non-ground point grid and removed from the plane grid, and the remaining grid is marked as a ground grid.

6. The method for ground segmentation based on adaptive grid division and intensity information as claimed in claim 5, characterized in that: Determine the initial seed grid, including: Select grids within the set distance range of the laser radar from the plane grid as candidate grids for the initial seed grid, and calculate the average height of the point cloud in each candidate grid , if the average height Greater than the height information threshold , it is determined not to be a ground grid. After removing all ground grids, the remaining candidate grids are calculated for flatness respectively, and arranged in ascending order of flatness, and the grid with the smallest flatness is selected as the initial seed grid.

7. The ground segmentation method based on adaptive grid division and intensity information according to claim 1, characterized in that: Dynamically grouping the ground grid based on the intensity information to obtain intensity grid categories, including: Traverse all ground grids and calculate the average intensity value of each ground grid; Sort the average intensity values ​​of all ground grids from small to large to get an ordered sequence: Traverse the ordered sequence of intensity values, starting from the first ground grid with the smallest average intensity value, and group together the ground grids whose average intensity difference between adjacent ground grids is less than the intensity difference threshold. If the intensity difference is greater than the intensity difference threshold, start a new grouping. All ground grids are grouped and each group is defined as an intensity grid class.

8. The ground segmentation method based on adaptive grid division and intensity information as claimed in claim 7, characterized in that: Average intensity value ,include: ,in, For the grid i The intensity value of each point, N is the total number of point clouds in the grid.

9. The ground segmentation method based on adaptive grid division and intensity information as claimed in claim 1, characterized in that: Confidence correction is performed on the intensity grid categories to obtain ground points and non-ground points, including: The intensity grid category containing the largest number of point clouds is the high-confidence intensity grid. The point clouds contained in the high-confidence intensity grid are all ground points, and the remaining intensity grids are low-confidence intensity grids. For any low-confidence grid, if there are more than three high-confidence grids in the neighborhood of this low-confidence grid, this low-confidence grid will be corrected to a high-confidence grid, otherwise no correction will be performed; After confidence correction, the point clouds in the high-confidence grids are marked as ground point clouds, and the point clouds in the remaining grids are marked as non-ground point clouds.

10. A ground segmentation system based on adaptive grid division and intensity information, characterized in that: include: The point cloud acquisition unit is configured to: acquire laser radar point cloud data; The point cloud preprocessing unit is configured to: perform height and intensity preprocessing on the laser radar point cloud data, remove noise and invalid point clouds, and obtain preprocessed point cloud data; The grid division unit is configured to: perform adaptive grid division on the pre-processed point cloud data to obtain a plane grid; A grid region growing unit is configured to: perform grid region growing on the plane grid to obtain a ground grid; The grid dynamic grouping unit is configured to: dynamically group the ground grids based on the intensity information to obtain intensity grid categories; The grid confidence correction unit is configured to: perform confidence correction on the intensity grid category to obtain ground points and non-ground points.

Images