A ground filtering method and system based on lidar

Abstract

This invention relates to the field of testing and measurement, and provides a ground filtering method and system based on lidar. The lidar-based ground filtering method generates a lidar depth map, calculates the angular difference between two line beams in each column of the lidar depth map, and generates an angular disparity map of the same size as the lidar depth map. A label map of the same size is generated based on the angular disparity map, and the true ground point cloud in the label map is calculated and filtered out. The original lidar point cloud is then filtered out based on the label map and processed using the RANSAC algorithm to obtain the final ground point cloud. This enables rapid and accurate ground identification when dealing with complex terrain or changing scenes.

Description

Technical Field

[0001] This invention relates to the field of measurement and testing, and specifically to a ground filtering method and system based on lidar. Background Technology

[0002] In the daily operation of unmanned container trucks in ports, LiDAR is one of the main sensors used for obstacle avoidance. To effectively identify obstacles, the ground surface must first be thoroughly filtered before identifying non-ground point clouds. Traditional methods primarily use Singular Value Decomposition (SVD) or RANSAC for identification, which can identify the ground to some extent, but cannot handle scenarios such as uneven ground or changes in the operating environment. Furthermore, in actual port operations, the laser position angle often shifts due to varying container loads, increasing the accuracy and difficulty of ground identification.

[0003] Application CN202011179679.7, entitled "A Method and System for Removing Ground Based on LiDAR Point Cloud Data," combines LiDAR point cloud data with camera image data, using a transformation matrix to filter and remove ground features, thereby obtaining accurate ground data. However, the filtering method in "A Method and System for Removing Ground Based on LiDAR Point Cloud Data" requires effective use of image information to remove false detections of ground features on high platforms and rooftops within the detection range of a surround-view camera, which differs from the method in this application. Furthermore, the solution in this application employs multiple filtering processes, resulting in higher accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a ground filtering method and system based on lidar, which aims to improve the problem that traditional lidar cannot cope with some complex terrain or scene changes when identifying ground targets.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a ground filtering method based on lidar, comprising:

[0006] Power on and turn on the lidar;

[0007] Based on the scanning feedback of the lidar, a corresponding lidar depth map is generated, the depth values ​​in the lidar depth map are initialized, the depth values ​​between each line beam and the lidar are calculated and reassigned;

[0008] Calculate the angle difference between the two beams based on each column of the laser depth map, and generate an angle parallax map of the same size as the laser depth map.

[0009] A label map of the same size is generated based on the angle parallax map. In the initial state, the first label value of each column of the label map is assigned a fixed value and set as the undetermined point cloud on the ground. The point cloud of each line beam of the lidar at the ground position is verified to form a candidate point cloud. The undetermined point cloud is used as the starting point to perform a breadth search of the field and form a neighborhood point cloud.

[0010] If the candidate point cloud or the neighboring point cloud has a label value that matches the setting of the point cloud to be determined, then it is a real ground point cloud.

[0011] The original laser point cloud was selected based on the label image and then filtered using the RANSAC algorithm to obtain the final ground point cloud.

[0012] Preferably, the vertical resolution of the lidar is L, the horizontal resolution is H, the number of laser beams is S, and the point cloud data coordinates of the lidar are (x, y, z).

[0013] The length of the laser depth map is: rows = S + 1;

[0014] Width is: columns = 360 ÷ L;

[0015] The depth value is: depth = sqrt(x×x + y×y + z×z);

[0016] VA=atan2(z, sqrt(x×x+y×y)); HA=atan2(x, y);

[0017] RI=(VA-VAm)÷L; CI=(HA÷H)+columns÷2.0;

[0018] Where VA is the vertical angle of the lidar, VAm is the minimum angle value that the first beam of the lidar can illuminate, HA is the horizontal angle of the lidar, RI is the length index value of the lidar, and CI is the width index value of the lidar.

[0019] Preferably, when generating the laser depth map, NaN filtering is performed. A depth threshold is manually set to filter out point clouds with depth values ​​less than the depth threshold. Furthermore, false point clouds generated by the lidar and point clouds beyond 100 meters are filtered. The filtered point clouds are assigned a value of 0. A 0-value point is selected, and a sliding window is set to match point clouds adjacent to the 0-value point. The depth difference between the adjacent point cloud and the 0-value point is calculated. If the depth difference meets the depth threshold, it is set as a candidate value. This process is repeated to filter all 0-value points. A weighted average is calculated for all candidate values, all 0-values ​​are assigned a value, and the laser depth map is generated.

[0020] Preferably, let P be the angle difference between two line beams calculated for each column of the laser depth map; and let r represent any row and c represent any column in the laser depth map; let A and B represent two different point clouds of the laser radar on the ground, and let the location of the laser radar be the origin 0 of the coordinate system to establish a coordinate system of X-axis and Z-axis, let E be the point of point cloud A perpendicular to the Z-axis, let D be the point of point cloud B perpendicular to the X-axis, and let Q be the point where the line connecting AE and BD intersects, forming a right triangle ABQ with Q as the right angle;

[0021] α = ∠AOD; β = ∠BOD;

[0022] P = atan2(|BQ|, |AQ|);

[0023] |BQ|=|QD|-|BD|=depth r-1，c sinα-depth r，c sinβ;

[0024] |AQ|=|QE|-|AE|=depth r-1，c cosα-depth r，c cosβ;

[0025] All the obtained angle differences are filtered and smoothed using image convolution filtering.

[0026] Preferably, the line bundles of the angle disparity map are traversed, and when the obtained angle difference is less than the maximum angle difference, it is set as a candidate point cloud of the ground;

[0027] Let IH be the height of the lidar above the ground after installation. Then the depth value of the lidar is:

[0028] θ=∠AOE; depth=IH÷cosθ;

[0029] The difference between the depth value of all candidate point clouds and the current depth is calculated and compared with the set verification threshold. Candidate point clouds that meet the verification threshold are set as the real ground point clouds.

[0030] Preferably, the value of the verification threshold increases with the increase of the depth value, and the verification threshold is set with a floating threshold. When the lidar adds a beam, the verification threshold increases by one of the floating thresholds.

[0031] Preferably, the angle difference threshold between the neighboring point cloud and the current point cloud is DT, the bottom surface movement angle value of the initial ground similarity threshold is GRA, the angle ratio of the floating threshold is AR, and the row value of the current traversal is CR.

[0032] When DT < GRA + AR × CR,

[0033] then the neighborhood point cloud conforms to the alternative point cloud.

[0034] Preferably, when "filtering the original laser point cloud according to the label map" is completed,

[0035] Three of the original laser point clouds are selected within a fixed range according to the label map, the distance between each of the original laser point clouds is the same, a minimum distance is artificially set between two adjacent original laser point clouds, and a plane distance threshold is artificially set;

[0036] When the distance between two adjacent original laser point clouds is less than the minimum distance, re-selection is performed;

[0037] The three obtained original laser point clouds are subjected to plane fitting to obtain a fitting plane. Among multiple selections of the three original laser point clouds, the plane point cloud with the most selections is set as the plane coefficient. Combining the plane coefficient, the plane distance between all point clouds in the label map and the fitting plane is calculated in sequence;

[0038] When the plane distance is less than the plane distance threshold, the point cloud is the final ground point cloud.

[0039] Preferably, when performing the plane fitting, the centroid height of the original laser point cloud is compared with the installation height of the lidar, and the centroid height is set as HC;

[0040] When HC - IH > 0.5 m, re-plane fitting is performed;

[0041] The normal vector angle of the fitting plane is calculated. When the normal vector angle is less than 85 degrees, it is determined that the search for the fitting plane fails in this time, and the fitting plane is filtered.

[0042] A ground filtering system based on lidar is also provided, including:

[0043] A lidar for collecting the point cloud data of the lidar; and

[0044] A processor, which is data-connected to the lidar and performs the following operations based on the point cloud data of the lidar:

[0045] A coordinate system with the LiDAR as the origin is constructed. A laser depth map is constructed using the LiDAR beam indices and calculated values, and the NaN values ​​in the LiDAR data are filtered out. False point clouds acquired by the LiDAR and point clouds beyond 100 meters are reassigned. The angle difference between two beams is calculated based on each column of the laser depth map to construct an angle disparity map of the same size. Image convolution filtering is performed to smooth the data in the angle disparity map. The LiDAR beams in the angle disparity map are traversed, and the point cloud data in the surrounding grid are verified to select candidate point clouds. These candidate point clouds are then filtered or confirmed as real ground point clouds, generating a label map of the same size as the angle disparity map. A fixed range is selected within the label map, and three original laser point clouds are randomly selected from this range to construct a preliminary fitting plane. The planar distances between all point clouds and the fitting plane are calculated, and it is determined whether the conditions for plane fitting are met. If they are met, the point cloud is the final ground point cloud; otherwise, it is filtered.

[0046] By adopting the above technical solution, the present invention has the following advantages compared with the prior art:

[0047] 1. Using LiDAR, we construct a LiDAR depth map, angular parallax map, and label map. We filter out some NaN values ​​and process the point clouds that are far away and the false point clouds formed by the nearby points using the RANSAC algorithm to obtain the final ground point cloud. This is to ensure that we can still obtain accurate ground point clouds in the process of complex road surface or scene changes. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the lidar in the lidar-based ground filtering method described in this invention;

[0049] Figure 2 This is a coordinate system diagram established with the lidar as the origin in the ground filtering method based on lidar described in this invention.

[0050] Explanation of reference numerals in the attached figures:

[0051] 10. LiDAR; 20. Wire harness. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Furthermore, it should be noted that:

[0053] The terms “upper,” “lower,” “left,” “right,” “vertical,” “horizontal,” “inner,” and “outer” are based on the orientation or positional relationship shown in the accompanying drawings and are used merely for the convenience of describing the present invention and simplifying the description. They are not intended to indicate or imply that the device or element of the present invention must have a specific orientation and therefore should not be construed as a limitation of the present invention.

[0054] When an element is referred to as being "fixed to," "set on," or "contained on" another element, it can be directly on or indirectly on that other element. When an element is referred to as being "connected to," it can be directly connected to or indirectly connected to that other element.

[0055] Unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0056] Example

[0057] Please refer to Figure 1 As shown in the accompanying drawings and specific embodiments, the ground filtering method and system based on lidar described in this invention will be further explained and described below.

[0058] This implementation provides a ground filtering method based on LiDAR, including: powering on the LiDAR; generating a corresponding laser depth map based on the scanning feedback of the LiDAR 10, initializing the depth values ​​in the laser depth map, calculating the depth values ​​between each beam and the LiDAR and reassigning them; calculating the angle difference between two beams 20 based on each column of the laser depth map, generating an angle disparity map of the same size as the laser depth map; generating a label map of the same size based on the angle disparity map, in the initial state of the label map, assigning a fixed value to the first label value of each column and setting it as the undetermined point cloud of the ground, and verifying the point cloud of each beam of the LiDAR 10 at the ground position to form a candidate point cloud, performing a breadth search of the neighborhood using the undetermined point cloud as the starting point, and forming a neighboring point cloud; if the candidate point cloud or the neighboring point cloud has a label value that matches the set value of the undetermined point cloud, it is the real ground point cloud; filtering out the original laser point cloud based on the label map, and filtering it through the RANSAC algorithm to obtain the final ground point cloud.

[0059] The lidar 10 operates using a mechanical laser scanning method, such as 16-line, 32-line, or 64-line laser devices. The laser installation position needs to ensure it is roughly horizontal, and the beam can fully reach the ground. To facilitate understanding of the above method, taking a 16-line lidar as an example, the highest beam is essentially in the air and does not affect ground detection. Therefore, a new beam data is constructed, with beam 20 data identical to the 16th beam data, but beam 20 is moved by 2°.

[0060] Specifically, the vertical resolution of the lidar is L, where L = 2°, the horizontal resolution is H, where H = 0.2°, the number of laser beams is S, where S = 16, and the point cloud data coordinates of the lidar are (x, y, z), which is the origin of the coordinate system.

[0061] The length of the laser depth map is: rows = S + 1;

[0062] Width is: columns = 360 ÷ L;

[0063] The depth value is: depth = sqrt(x×x + y×y + z×z);

[0064] VA=atan2(z, sqrt(x×x+y×y)); HA=atan2(x, y);

[0065] RI=(VA-VAm)÷L; CI=(HA÷H)+columns÷2.0;

[0066] Where VA is the vertical angle of the lidar, HA is the horizontal angle of the lidar, RI is the length index value of the lidar, CI is the width index value of the lidar, and VAm is the minimum angle value that the first beam of the lidar can illuminate.

[0067] When generating laser depth maps, due to the inherent characteristics of lasers, a large number of NaN values ​​(NaN is an abbreviation for Not a Number. NaN is used to handle errors that occur in calculations, such as dividing 0.0 by 0.0 or taking the square root of a negative number) are present. Therefore, NaN filtering is required. A depth threshold of 0.2 meters is manually set to filter out point clouds with depths smaller than this threshold. Furthermore, false point clouds generated by the lidar and point clouds beyond 100 meters are also filtered out.

[0068] When generating the laser depth map, the aforementioned processing results in a large number of zero values. To compensate for the significant differences caused by this setting, appropriate smoothing is required. Specifically, the filtered point cloud is assigned a zero value. A zero-value point is selected as the current point, and a sliding window is set to match the point clouds adjacent to the zero value. This involves matching each point in the left window of the current point with the point in the right window, for example, forming 5 groups of windows, each with 5 pairs of data. The depth difference between the left and right values ​​in each group of windows is calculated. When the depth difference meets a depth threshold (e.g., 1m), the point cloud is included as a candidate value. This process is repeated to filter all zero-value points. A weighted average is then applied to all candidate values ​​(the weighting value can be manually set). This weighted average ensures that the closer two point clouds are, the smaller the difference becomes. After assigning values ​​to all zero values, a laser depth map with 17 rows is generated.

[0069] Furthermore, such as Figure 2 As shown, let P be the angle difference between two line beams calculated in each column of the laser depth map; and let r represent any row and c represent any column in the laser depth map; let A and B represent two different point clouds of the lidar on the ground, and let the location of the lidar be the origin 0 of the coordinate system to establish a coordinate system with the X-axis and Z-axis. The point of point cloud A that is perpendicular to the Z-axis is E, the point of point cloud B that is perpendicular to the X-axis is D, and the point where the line connecting AE and BD intersects is Q, forming a right triangle ABQ with Q as the right angle;

[0070] α = ∠AOD; β = ∠BOD;

[0071] P = atan2(|BQ|, |AQ|);

[0072] |BQ|=|QD|-|BD|=depth r-1，c sinα-depth r，c sinβ;

[0073] |AQ|=|QE|-|AE|=depth r-1，c cosα-depth r，c cosβ;

[0074] Using the formula above, the corresponding values ​​for the selected coordinate points, as well as the values ​​for arbitrarily selected points A and B, can be calculated. Specifically, r and r-1 represent two adjacent line beams. By taking values ​​from any row and column in the laser depth map, such as the value in the r-th row and c-th column, the pixel values ​​in the corresponding row and column are obtained, thus obtaining the required angle difference P. When the laser has 16 lines, due to the expansion processing of the laser depth map, the actual calculated angle difference is 16. The 17-row laser depth map mentioned above will have an extra row, which can be assigned a value of 0.

[0075] Similarly, the obtained angle disparity map also needs to be filtered and smoothed because some point cloud values ​​in the calculated angle disparity map differ significantly. To avoid outliers having high similarity to subsequent comparison points, image convolution filtering is used. Specifically, the Savitzky-Golay filter is employed. This filter allows for the selection of different window widths to smooth the data. The window size can be adjusted according to the actual laser scene data for filtering. This completes the smoothing process of all acquired angle differences using image convolution filtering.

[0076] Furthermore, a label map of the same size needs to be generated based on the angle disparity map. After generating the label map, the initial value is set to 0, while the first label value of each column is initially 1, assuming it to be a point cloud on the ground. Therefore, for a 16-line LiDAR, under relatively flat ground conditions, the angle difference between adjacent points in the angle disparity map is not significant. If it is a non-ground point cloud, the angle between the two will be larger; for example, the difference between two points in a vertical column on a wall will be 90°.

[0077] Therefore, the line bundles in the angle disparity map are traversed, starting with the nearest line bundle in each column. When the obtained angle difference is less than the maximum angle difference, it is set as a candidate point cloud for the ground. The maximum angle difference can be set to 30°. Simultaneously, the approximate location of the original points is needed. The ground location of each line bundle point cloud is estimated a priori. Let the height of the LiDAR above the ground after installation be IH, which is the length between O and E in the coordinate system. Then the depth value of the LiDAR is:

[0078] θ=∠AOE; depth=IH÷cosθ;

[0079] The depth value of the LiDAR can be calculated using the formula described above. The difference between the depth value of all candidate point clouds and the current depth is calculated and compared with a set verification threshold. Candidate point clouds that meet the verification threshold are set as the real ground point clouds.

[0080] Furthermore, the verification threshold increases with the depth value. A floating threshold is set for the verification threshold; when a new laser beam is added to the lidar system, the verification threshold increases by one floating threshold. Let the initial verification threshold be 1 meter, and the floating threshold be 0.2 meters (the floating threshold can be optimized based on actual results). The floating threshold means that as the number of laser beams increases, the threshold increases by 0.2 meters for each additional laser beam.

[0081] Using breadth - first search, search and spread using the first ground point cloud of each column above. Find the neighborhood of the current ground point cloud. For example, the neighborhood point clouds are the 4 neighborhood point clouds on the left, right, above, and below respectively, or set to 8 neighborhood point clouds, which can be selected according to the effect of the lidar point cloud. Calculate the angle difference threshold between the neighborhood point cloud and the current point cloud. The actual characteristics of the laser for distant objects and nearby objects are different. The point cloud of nearby objects is denser and the point cloud accuracy is higher, while the points of distant objects are sparser and are also prone to deformation. Therefore, for the difference between distant point clouds and nearby point clouds, the adjustment of the adaptive neighbor coefficient can be made. For example, different thresholds can be used for adjacent point clouds according to the position of the current point cloud.

[0082] Specifically, let the angle difference threshold between the neighborhood point cloud and the current point cloud be DT, the bottom surface movement angle value of the initial ground similarity threshold be GRA, GRA = 5°, the angle ratio of the floating threshold be AR, AR = 0.2°, and the current traversed row value be CR. The system in the ground filtering method based on lidar will plan the lidar depth map, angle disparity map, and label map as grids with multiple rows and columns for traversing one by one, or traversing the whole row or the whole column;

[0083] When DT < GRA + AR×CR,

[0084] Then the neighborhood point cloud meets the alternative point cloud, and the neighborhood point cloud is screened. In addition, for the neighborhood in the same row, the depth difference needs to be added for limitation. The verification threshold can be used, or it can be reset by oneself, and the set value is adjusted according to the laser characteristics.

[0085] In this way, after traversing each column in the label map, the label map can be generated through the above steps. The ground point cloud is the one with a value of 1 in the label.

[0086] In this embodiment, it is also necessary to filter the ground through the ransac algorithm. Specifically, when "selecting the original lidar point cloud according to the label map" is completed:

[0087] First, select the lidar point cloud according to the label map. That is, through the label map, select the original lidar point cloud corresponding to the position. Here, nan value filtering needs to be performed, and the point cloud too close to the laser (filtered for less than 0.2 meters) is removed. The target too far away (the point cloud exceeding 100 meters is filtered).

[0088] Second, ransac bottom surface filtering, that is, further filtering the lidar point cloud obtained in the first step. The processing steps are as follows:

[0089] 1. Select three original lidar point clouds within a fixed range according to the label map; in order to avoid the sudden change of the point cloud value in the distant part and unreasonable point cloud data, limit by a fixed range to ensure the uniformity of the point cloud. Hierarchical limitation can be made on the area in the label map. The specific selection requirements are as follows:

[0090] The fixed range can be a point cloud within 21 meters. The spacing between each laser-generated point cloud is the same; for example, selecting one point cloud every 7 meters will yield three laser-generated point clouds.

[0091] A minimum distance is manually set between two adjacent raw laser point clouds, for example, a minimum distance of 2 meters, to prevent the two point clouds from being too close. If a point cloud is found to be less than 2 meters away, it will be reselected.

[0092] Finally, a planar distance threshold is set manually, which can be 0.3 meters.

[0093] 2. Obtain planar parameters. Specifically, perform planar fitting on the three original laser point clouds to obtain the fitted plane, and filter the fitted planes that do not meet the following requirements.

[0094] When performing plane fitting, the centroid height of the original laser point cloud is compared with the installation height of the lidar. Let the centroid height be HC. Therefore, when HC-IH > 0.5 meters, plane fitting is performed again.

[0095] The normal vector angle of the fitted plane is calculated. If the normal vector angle is less than 85 degrees, the fitted plane search is considered to have failed and the fitted plane is filtered out.

[0096] By selecting from the three original laser point clouds multiple times, the planar point cloud that is selected most frequently is set as the planar coefficient.

[0097] 3. Ground Filtering Process. Specifically, using the plane coefficients obtained above, the planar distance between each point cloud in the label image and the fitted plane is calculated sequentially; when the planar distance is less than the planar distance threshold (i.e., less than 0.3 meters), the point cloud is considered the final ground point cloud.

[0098] Therefore, the laser point cloud filtering method combining laser depth maps, angular parallax maps, tag maps, and Ransac can effectively handle problems such as slight slopes, heavy-duty containers, and minor laser position fluctuations. Furthermore, the ground filtering method based on lidar can filter ground point clouds in real-time processing of unmanned container trucks in ports, accurately obtaining planar parameters and aiding in the estimation of the ground environment.

[0099] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A ground-based filtering method based on lidar, characterized in that, include: Power on and turn on the lidar; Based on the scanning feedback of the lidar, a corresponding lidar depth map is generated, the depth values ​​in the lidar depth map are initialized, the depth values ​​between each line beam and the lidar are calculated and reassigned; Calculate the angle difference between the two beams based on each column of the laser depth map, and generate an angle parallax map of the same size as the laser depth map. A label map of the same size is generated based on the angle parallax map. In the initial state, the first label value of each column of the label map is assigned a fixed value and set as the undetermined point cloud on the ground. The point cloud of each line beam of the lidar at the ground position is verified to form a candidate point cloud. The undetermined point cloud is used as the starting point to perform a breadth search of the field and form a neighborhood point cloud. If the candidate point cloud or the neighboring point cloud has a label value that matches the setting of the point cloud to be determined, then it is a real ground point cloud. The original laser point cloud is selected based on the label map, and a plane is fitted using the original laser point cloud. The ground representations below the ground threshold are filtered out using the rgnsac algorithm to obtain the final ground point cloud.

2. The ground-based filtering method based on lidar according to claim 1, characterized in that: Let the vertical resolution of the lidar be L, the horizontal resolution be H, the number of laser beams be S, and the point cloud data coordinates of the lidar be (x, y, z). The length of the laser depth map is: rows = S + 1; Width is: columns = 360 ÷ L; The depth value is: depth = sqrt(x×x + y×y + z×z); VA=atan2(z, sqrt(x×x+y×y)); HA=atan2(x, y); RI=(VA-VAm)÷L; CI=(HA÷H)+columns÷2.0; Where VA is the vertical angle of the lidar, VAm is the minimum angle value that the first beam of the lidar can illuminate, HA is the horizontal angle of the lidar, RI is the length index value of the lidar, and CI is the width index value of the lidar.

3. The ground-based filtering method based on lidar according to claim 1, characterized in that: When the laser depth map is generated, it is filtered for Nan values. A depth threshold is manually set to filter out point clouds with depth values ​​less than the depth threshold. In addition, false point clouds generated by the lidar and point clouds beyond 100 meters are filtered out. The filtered point cloud is assigned a value of 0. A point with a value of 0 is selected and a sliding window is set to match the point cloud adjacent to the point with a value of 0. The depth difference between the adjacent point cloud and the point with a value of 0 is calculated. When the depth difference meets the depth threshold, it is set as a candidate value. This process is repeated to filter all points with a value of 0. A weighted average is calculated for all candidate values, all zero values ​​are assigned a value, and the laser depth map is generated.

4. The ground-based filtering method based on lidar according to claim 2, characterized in that: Let P be the angle difference between two line beams calculated in each column of the laser depth map, and let r represent any row and c represent any column in the laser depth map; let A and B represent two different point clouds of the lidar on the ground, and let the location of the lidar be the origin 0 of the coordinate system to establish a coordinate system with the X-axis and Z-axis. Let E be the point of point cloud A that is perpendicular to the Z-axis, let D be the point of point cloud B that is perpendicular to the X-axis, and let Q be the point where the line connecting AE and BD intersects, forming a right triangle ABQ with Q as the right angle; α = ∠AOD; β = ∠BOD; P = atan2(|BQ|, |AQ|); |BQ|=|QD|-|BD|=depth r-1，c sinα-depth r，c sinβ; |AQ|=|QE|-|AE|=depth r-1，c cosα-depth r，c cosβ; All the obtained angle differences are filtered and smoothed using image convolution filtering.

5. The ground-based filtering method based on lidar according to claim 4, characterized in that: The line bundles of the angle disparity map are traversed, and when the obtained angle difference is less than the maximum angle difference, it is set as a candidate point cloud of the ground. Let IH be the height of the lidar above the ground after installation. Then the depth value of the lidar is: θ=∠AOE; depth=IH÷cosθ; The difference between the depth value of all candidate point clouds and the current depth is calculated and compared with the set verification threshold. Candidate point clouds that meet the verification threshold are set as the real ground point clouds.

6. The ground-based filtering method based on lidar according to claim 5, characterized in that: The value of the verification threshold increases with the increase of the depth value. The verification threshold is set with a floating threshold. When the lidar adds a beam, the verification threshold increases by one of the floating thresholds.

7. The ground-based filtering method based on lidar according to claim 6, characterized in that: Let DT be the threshold for the angle difference between the neighboring point cloud and the current point cloud, GRA be the bottom surface movement angle value of the initial ground similarity threshold, AR be the angle ratio of the floating threshold, and CR be the row value of the current traversal. When DT < GRA + AR × CR Then the neighborhood point cloud conforms to the candidate point cloud.

8. The ground-based filtering method based on lidar according to claim 5, characterized in that: When the task of "filtering out the original laser point cloud based on the label image" is completed... Three original laser point clouds are selected within a fixed range based on the label image. The spacing between each original laser point cloud is the same. A minimum distance is set between two adjacent original laser point clouds, and a planar distance threshold is set manually. If two adjacent original laser point clouds are less than the minimum distance, then a new selection is performed; The three original laser point clouds are fitted with a plane to obtain a fitting plane. By selecting the three original laser point clouds multiple times, the plane point cloud with the most selections is set as the plane coefficient. Based on the plane coefficient, the plane distance between all point clouds in the label image and the fitting plane is calculated in turn. When the planar distance is less than the planar distance threshold, the point cloud is the final ground point cloud.

9. The ground-based filtering method based on lidar according to claim 8, characterized in that: When performing the plane fitting, the centroid height of the original laser point cloud is compared with the installation height of the lidar, and the centroid height is set as HC; When HC-IH > 0.5 m, perform plane fitting again; The normal vector angle of the fitted plane is calculated. If the normal vector angle is less than 85 degrees, the fitted plane search is considered to have failed and the fitted plane is filtered out.

10. A ground-based filtration system based on lidar, characterized in that, include: LiDAR, used to collect point cloud data; and A processor, which is connected to the lidar data network, performs the following operations based on the lidar's point cloud data: Construct a coordinate system with the lidar as the origin; A laser depth map is constructed using the line index of the lidar and the calculated values. The nano values ​​in the lidar are filtered out, and false point clouds and point clouds beyond 100 meters are reassigned. The angle difference between the two beams is calculated based on each column of the laser depth map to construct an angle disparity map of the same size. The data in the angle disparity map is then smoothed by performing image convolution filtering through a filter. The laser radar beams in the angle parallax map are traversed and verified with the point cloud data in the grid of the field. Candidate point clouds are selected and verified for filtering or confirmation as real ground point clouds. A label map of the same size as the angle parallax map is generated. A fixed range is selected from the labeled image, and three original laser point clouds are randomly selected from the fixed range to construct a preliminary fitting plane; the planar distance between all point clouds and the fitting plane is calculated, and it is determined whether the conditions for plane fitting are met. If they are met, the point cloud is the final ground point cloud; if not, it is filtered.

Images