Vehicle LiDAR Point Cloud Clustering Algorithm Based on Curvature Information for Indoor Environments

Abstract

The present invention provides a vehicle-mounted lidar point cloud clustering algorithm based on curvature information for indoor environments, which includes sequentially performing steps such as data preprocessing, ground point segmentation, true and false breakpoint determination, point cloud curvature calculation, and breadth-first search. The data preprocessing stores the point cloud data and subsequent processing results as a two-dimensional matrix; the ground point segmentation performs the segmentation of ground points in the point cloud; the true and false breakpoint determination performs the determination of wall breakpoints in the point cloud to distinguish true wall breakpoints from false breakpoints caused by blocked vision; the point cloud curvature calculation performs the curvature calculation of discrete point clouds based on the curvature calculation method in polar coordinates; the breadth-first search performs category determination according to the numerical comparison result of the point cloud curvature and the threshold parameter, and performs two-dimensional breadth search. The present invention can achieve the lidar point cloud clustering segmentation of different wall features and the wall features and corner features in indoor scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of lidar point cloud processing, and particularly relates to the clustering processing of vehicle-mounted lidar point clouds based on curvature information for indoor environments. Background Art

[0002] The advantage of lidar over traditional distance sensing means is that the distance information contained in the acquired lidar point cloud is richer and more accurate, but it also means a larger amount of data and a more discrete distribution. As one of the existing lidar point cloud processing methods, point cloud segmentation and clustering can sort out the chaotic point cloud information, facilitating the extraction of effective point cloud features for further processing. Currently, the commonly used point cloud segmentation algorithms mainly include DBSCAN, K-means, and Euclidean clustering, etc. The central idea of the DBSCAN algorithm is to divide the point cloud data set into multiple clusters according to the density between each point cloud. It has the advantages of simple algorithm principle, easy implementation, and insensitivity to noise. However, with the increase in the amount of lidar point cloud data, problems such as slower clustering convergence speed will occur, and when the spatial density is uneven, missed detections and false detections will occur. The central idea of the K-means algorithm is to repeatedly cluster the point cloud data into K clusters based on the Euclidean distance from the point cloud data to K cluster centers. It has the advantages of simple algorithm principle, easy implementation, fast operation, and good clustering effect. However, it is sensitive to noise and outliers, and has a high time complexity. The selection of K will also affect the clustering effect. At the same time, both the DBSCAN and K-means algorithms only process the first-order distance information of the point cloud, dividing the point cloud into geometric bodies that are spatially separated from each other, but they cannot segment different geometric surfaces of the same geometric body, which greatly limits their applications. Summary of the Invention

[0003] In view of this, in order to solve the above technical problems existing in the art, the present invention provides a vehicle-mounted lidar point cloud clustering algorithm based on curvature information for indoor environments, which specifically includes the following steps:

[0004] Step 1: Determine the size of a two-dimensional matrix based on the horizontal and vertical viewing angles of the lidar. Based on the distances and azimuths of each lidar point relative to the lidar in the horizontal and vertical directions, determine the coordinates of the corresponding point clouds in the two-dimensional matrix, and perform matrix filling on the two-dimensional matrix based on the three-dimensional coordinates of each point and its distance relative to the lidar according to the coordinates.

[0005] Step 2: Calculate the elevation angle and the elevation angle difference in the vertical direction for adjacent lidar points, and mark the points where both the elevation angle and the elevation angle difference are less than the corresponding thresholds as ground points.

[0006] Step 3: Calculate the point cloud spacing for adjacent laser points, extract the laser points with a spacing greater than the breakpoint determination threshold as pseudo-breakpoints, and combine the relative positions to determine whether each pseudo-breakpoint is a true breakpoint or a false breakpoint; where a true breakpoint represents a discontinuity in the horizontal distribution of the point cloud caused by the edge of the wall, and a false breakpoint represents a discontinuity in the horizontal distribution of the point cloud caused by the obstruction of the lidar's field of view.

[0007] Step 4: Calculate the curvature in the horizontal direction at the position of each laser point using the point cloud polar coordinate curvature calculation formula, and store the results as a two-dimensional matrix of point cloud curvature based on the coordinates of the point cloud in the two-dimensional matrix.

[0008] Step 5: Determine whether the category of each laser point belongs to a pseudo-breakpoint or a ground point, and combine the two-dimensional matrix of point cloud curvature, the corresponding curvature threshold, and the true / false breakpoint judgment category to determine whether it belongs to a planar point or a wall edge point on the wall; perform a breadth-first search clustering on all points based on the judged category results, set the same number for the same category of points and store them as a two-dimensional matrix of clustering results, so as to realize the recognition of the ground and different walls in the indoor environment.

[0009] Further, the size of the two-dimensional matrix in Step 1 is specifically determined based on the following formula:

[0010] N = Θ / dθ

[0011] H = Φ / dφ

[0012] In the formula, N and H are the sizes of the horizontal and vertical dimensions of the two-dimensional matrix respectively, Θ and dθ are the horizontal viewing angle and horizontal angular resolution of the lidar respectively, and Φ and dф are the vertical viewing angle and vertical resolution of the lidar respectively.

[0013] The coordinates of each point cloud in the two-dimensional matrix are specifically determined based on the following formula:

[0014] Matx = θ / dθ

[0015] Maty = φ / dφ

[0016] In the formula, Matx is the abscissa of a certain point in the two-dimensional matrix, Maty is the ordinate of a certain point in the two-dimensional matrix, θ is the horizontal angle of a certain point relative to the lidar, and ф is the pitch angle of a certain point relative to the lidar. The specific calculation formula is as follows:

[0017] θ = atan2(y, x)

[0018]

[0019] In the formula, x, y, and z correspond to the three-dimensional coordinates of the laser point relative to the lidar respectively.

[0020] Further, for the elevation angle α of the i-th point in Step 2 iAnd the elevation angle difference Δα with the adjacent point directly above i is obtained based on the following formula:

[0021]

[0022] Δα i = α i+1 - α i

[0023] In the formula, the subscripts i and i + 1 represent the i-th point and its adjacent (i + 1)-th point directly above respectively;

[0024] For α i and Δα i Points that are both less than the corresponding thresholds are marked as ground points and stored as a two-dimensional matrix.

[0025] Furthermore, the determination of true and false breakpoints in step three specifically includes:

[0026] First, distinguish between horizontal breakpoints and vertical breakpoints. If there are multiple consecutive pseudo-breakpoints in the horizontal direction, it is determined as a horizontal breakpoint; if it is not a horizontal breakpoint and the distance between two adjacent points in the horizontal direction is greater than the distance threshold, it is determined as a vertical breakpoint;

[0027] Then, determine the true and false of the breakpoints. All horizontal breakpoints are determined as true breakpoints, and among the vertical breakpoints, those within a specific range from the lidar are determined as true breakpoints, and those outside the range are determined as false breakpoints.

[0028] Furthermore, in step four, calculating the curvature of each point in the horizontal direction is first based on the polar radius ρ(θ, ф) in the polar coordinate system, and its discretized first derivative ρ’(θ, ф) and second derivative ρ”(θ, ф) are calculated using the difference quotient:

[0029]

[0030]

[0031] Δθ = mdθ

[0032] In the formula, ρ(θ + Δθ, ф) and ρ(θ - Δθ, ф) are calculated using the polar radii of the adjacent n points:

[0033]

[0034]

[0035] Both of the above parameters m and n are selected as appropriate integer values according to actual requirements;

[0036] Then perform the following calculations to first obtain the discrete curvature calculation result κ(θ, ф), and then perform distance normalization using the average distance dis(θ, ф) in the horizontal direction to obtain the normalized curvature result.

[0037]

[0038] dis(θ,φ)=(ρ(θ+Δθ,φ)+ρ(θ,φ)+ρ(θ-Δθ,φ)) / 3

[0039]

[0040] Further, the judgment rules for the categories to which each laser point belongs in step five, in addition to the judgment of ground points, quasi-break points, and true and false break points in steps two to four, also include:

[0041] Except for quasi-break points and ground points, points with curvature greater than the curvature threshold and true break points in the true and false break point determination are determined as wall horizontal edge points;

[0042] Except for quasi-break points and ground points, points with curvature less than the curvature threshold and false break points in the true and false break point determination are determined as wall upper plane points.

[0043] Further, the implementation of breadth-first search clustering in step five specifically includes the following steps:

[0044] (1) Select a point from the unclustered points and use it as the first point in the point cloud sequence to be searched;

[0045] (2) Take out the first point in the point cloud sequence to be searched as the search center and determine the category to which it belongs;

[0046] (3) Perform a rationality determination on the four points above, below, left, and right of the search center, and add the points determined to be reasonable to the point cloud sequence to be clustered;

[0047] (4) Take out the first point in the point cloud sequence to be clustered and determine its category. If it is the same as the category of the search center, add it to the queue to be searched. If it is different from the category of the search center, no clustering processing is performed. If all the points in the point cloud sequence to be clustered have been taken out, add the search center to the searched queue;

[0048] (5) If the sequence to be searched is not an empty set, return to step (2). If the queue to be searched is an empty set, the search process is closed, and all the points in the searched queue form a point cloud class, set a class number for it, and store the class number in the two-dimensional clustering result matrix;

[0049] (6) Terminate the search algorithm when all points have been clustered. If there are points that have not been clustered, return to step (1).

[0050] Further, the rationality judgment in step (3) includes:

[0051] Determine whether the four points above, below, left, and right of the search center are within the valid coordinate range. If the horizontal coordinate of a point exceeds the upper limit, assign the horizontal coordinate as 0; if the horizontal coordinate of a point is less than the lower limit, assign the horizontal coordinate as the total number of horizontal laser beam lines - 1; if the coordinate in the vertical direction exceeds the coordinate range, mark this point as a pseudo-break point; determine whether the point has been clustered before, and if so, skip this point.

[0052] Further, when using a single-line lidar to execute the method provided by the present invention, a one-dimensional array is used to replace the two-dimensional matrix form for point cloud processing and storage in each step, and the ground point marking step in step two is omitted.

[0053] The above-mentioned on-vehicle lidar point cloud clustering algorithm based on curvature information for indoor environments provided by the present invention, based on the elevation angle threshold and the true and false break point determination means, can effectively and accurately identify the ground points and walls in the environment. Combining the point cloud curvature calculated in the polar coordinate system and the breadth-first search clustering, it can quickly and clearly obtain the area range and edges of different walls, simplify the method process, improve the processing speed, and significantly reduce the calculation amount. This method is not only applicable to multi-line lidars, but also applicable to single-line lidars after simple adjustment, thus providing wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The flow chart of the method provided by the present invention:

[0055] Figure 2 It is a schematic diagram of a horizontal break point of a wall;

[0056] Figure 3 It is a schematic diagram of a vertical break point of a wall;

[0057] Figure 4 It is an indoor scene model diagram in an example based on the present invention

[0058] Figure 5 It is the clustering processing result for the indoor scene model in an example based on the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] The vehicle-mounted lidar point cloud clustering algorithm based on curvature information for indoor environments provided by the present invention has a specific process as Figure 1 shown, including the following steps:

[0061] Step 1: Determine the size of a two-dimensional matrix based on the horizontal and vertical perspectives of the lidar. Based on the distances and azimuths of each laser point relative to the lidar in the horizontal and vertical directions, determine the coordinates of the corresponding point clouds in the two-dimensional matrix, and perform matrix filling on the two-dimensional matrix based on the three-dimensional coordinates of each point and its distance relative to the lidar using the coordinates.

[0062] Step 2: Calculate the elevation angle and elevation angle difference in the vertical direction for adjacent laser points, and mark the points where both the elevation angle and elevation angle difference are less than the corresponding thresholds as ground points.

[0063] Step 3: Calculate the point cloud spacing for adjacent laser points, extract the laser points with a spacing greater than the breakpoint determination threshold as pseudo-breakpoints, and combine the relative positions to determine whether each pseudo-breakpoint is a true breakpoint or a false breakpoint; where a true breakpoint represents a discontinuity in the horizontal distribution of the point cloud caused by the edge of the wall, and a false breakpoint represents a discontinuity in the horizontal distribution of the point cloud caused by the obstruction of the lidar's field of view.

[0064] Step 4: Calculate the curvature in the horizontal direction at the position of each laser point using the point cloud polar coordinate curvature calculation formula, and store the results as a two-dimensional point cloud curvature matrix based on the coordinates of the point cloud in the two-dimensional matrix.

[0065] Step 5: Determine whether the category of each laser point belongs to a pseudo-breakpoint or a ground point, and combine the two-dimensional point cloud curvature matrix, the corresponding curvature threshold, and the true / false breakpoint determination to determine whether the category belongs to a planar point or a wall edge point on the wall; perform breadth-first search clustering on all points based on the determined category results, set the same number for the same category points and store them as a two-dimensional clustering result matrix, thereby realizing the recognition of the ground and different walls in the indoor environment.

[0066] In a preferred embodiment of the present invention, the size of the two-dimensional matrix in Step 1 is specifically determined based on the following formula:

[0067] N = Θ / dθ

[0068] H = Φ / dφ

[0069] In the formula, N and H are the sizes of the horizontal and vertical dimensions of the two-dimensional matrix respectively, Θ and dθ are the horizontal view angle and horizontal angular resolution of the lidar respectively, and Φ and dф are the vertical view angle and vertical resolution of the lidar respectively;

[0070] The coordinates of each point cloud in the two-dimensional matrix are specifically determined based on the following formula:

[0071] Matx = θ / dθ

[0072] Maty = φ / dφ

[0073] Wherein, Matx is the abscissa of a certain point in the two-dimensional matrix, Maty is the ordinate of a certain point in the two-dimensional matrix, θ is the horizontal angle of a certain point relative to the lidar, ф is the pitch angle of a certain point relative to the lidar, and the specific calculation formula is as follows:

[0074] θ = atan2(y, x)

[0075]

[0076] Wherein, x, y, and z respectively correspond to the three-dimensional coordinates of the laser point relative to the lidar.

[0077] Further, for the elevation angle α of the i-th point in step two i and the elevation angle difference Δα between it and the adjacent point directly above i are obtained based on the following formula:

[0078]

[0079] Δα i = α i+1 - α i

[0080] Wherein, the subscripts i and i + 1 respectively represent the i-th point and the (i + 1)-th point adjacent directly above it;

[0081] For α i and Δα i points that are both less than the corresponding thresholds are marked as ground points and stored as a two-dimensional matrix.

[0082] Figure 2 Shows the horizontal break point of the laser point cloud, which is formed by a horizontal gap perpendicular to the laser beam between two wall break points. Part of the laser beam shoots out from the gap without reflection, generating a pseudo-break point; Figure 3 Shows the longitudinal break point, which is formed by a longitudinal gap along the direction of the laser beam between two wall break points, and reflections of the same beam or adjacent beams on different walls. Therefore, in a preferred embodiment of the present invention, the determination of true and false break points in step three specifically includes:

[0083] First, distinguish between horizontal break points and longitudinal break points. If there are multiple consecutive pseudo-break points in the horizontal direction, it is determined as a horizontal break point; if it is not a horizontal break point and the distance between two adjacent points in the horizontal direction is greater than the distance threshold, it is determined as a longitudinal break point;

[0084] Then, determine the true and false of the break point. All horizontal break points are determined as true break points, and among the longitudinal break points, those within a specific range from the lidar are determined as true break points, and those outside the range are determined as false break points.

[0085] In a preferred embodiment of the present invention, in step four, calculating the curvature in the horizontal direction of each point is first based on the polar radius ρ(θ, ф) in the polar coordinate system, and using the difference quotient to calculate its discretized first derivative ρ’(θ, ф) and second derivative ρ”(θ, ф):

[0086]

[0087]

[0088] Δθ = mdθ

[0089] In the formula, ρ(θ + Δθ, ф) and ρ(θ - Δθ, ф) are calculated using the polar radii of the adjacent n points:

[0090]

[0091]

[0092] Both of the above parameters m and n are selected as suitable integer values according to actual requirements;

[0093] Then perform the following calculations to first obtain the discrete curvature calculation result κ(θ, ф), and then perform distance normalization using the average distance dis(θ, ф) in the horizontal direction to obtain the normalized curvature result

[0094]

[0095] dis(θ, φ) = (ρ(θ + Δθ, φ) + ρ(θ, φ) + ρ(θ - Δθ, φ)) / 3

[0096]

[0097] In a preferred embodiment of the present invention, the judgment rules for the categories of each laser point in step five, in addition to the judgment of ground points, quasi-break points, and true and false break points in steps two to four, further include:

[0098] Excluding quasi-break points and ground points, points with a curvature greater than the curvature threshold and true break points in the true and false break point determination are determined as wall horizontal edge points;

[0099] Excluding quasi-break points and ground points, points with a curvature less than the curvature threshold and false break points in the true and false break point determination are determined as wall upper plane points.

[0100] In a preferred embodiment of the present invention, performing breadth-first search clustering in step five specifically includes the following steps:

[0101] (1) Select a point from the unclustered points and use it as the first point of the point cloud sequence to be searched;

[0102] (2) Take out the first point in the point cloud sequence to be searched as the search center and determine the category to which it belongs;

[0103] (3) Perform a rationality determination on the four points above, below, left, and right of the search center, and add the points determined to be reasonable to the point cloud sequence to be clustered;

[0104] (4) Take out the first point in the point cloud sequence to be clustered and determine its category. If it is the same as the category of the search center, add it to the queue to be searched. If it is different from the category of the search center, no clustering process is performed. If all the points in the point cloud sequence to be clustered have been taken out, add the search center to the searched queue;

[0105] (5) If the sequence to be searched is not an empty set, return to step (2). If the queue to be searched is an empty set, the search process is closed. The points in the searched queue are formed into a point cloud class, a category number is set for it, and the category number is stored in the two-dimensional matrix of clustering results;

[0106] (6) Terminate the search algorithm when all points have been clustered. If there are points that have not been clustered, return to step (1).

[0107] In a preferred embodiment of the present invention, the rationality judgment in step (3) includes:

[0108] Determine whether the four points above, below, left, and right of the search center are within the valid coordinate range. If the horizontal coordinate of the point exceeds the upper limit, assign the horizontal coordinate to 0; if the horizontal coordinate of the point is less than the lower limit, assign the horizontal coordinate to the total number of horizontal laser beam lines - 1; if the coordinate in the vertical direction exceeds the coordinate range, mark this point as a pseudo-break point; determine whether the point has been clustered before, and if so, skip this point.

[0109] In a preferred embodiment of the present invention, when using a single-line lidar to execute the method provided by the present invention, a one-dimensional array is used to replace the form of a two-dimensional matrix for point cloud processing and storage in each step, and the ground point marking step in step two is omitted.

[0110] Figure 4 Shows an example based on the present invention. In the simulation environment, a complex indoor scene as shown in Figure 4 is established. By scanning the scene with a lidar and executing the method provided by the present invention, clear and accurate recognition results as shown in Figure 5 can be obtained for the ground and walls in the scene, and different walls and edges are well restored.

[0111] It should be understood that the sequence numbers of the steps in the embodiments of the present invention do not indicate the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0112] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for clustering the point cloud of an on-vehicle lidar based on curvature information for an indoor environment, characterized in that: Specifically, it includes the following steps: Step 1: Determine the size of the two-dimensional matrix based on the horizontal and vertical viewing angles of the lidar. Based on the distances and azimuths of each laser point relative to the lidar in the horizontal and vertical directions, determine the coordinates of the corresponding point clouds in the two-dimensional matrix. Perform matrix filling on the two-dimensional matrix for the three-dimensional coordinates of each point and its distance relative to the lidar based on the coordinates; Step 2: Calculate the elevation angle and elevation angle difference in the vertical direction for adjacent laser points, and mark the points where both the elevation angle and elevation angle difference are less than the corresponding thresholds as ground points; Step 3: Calculate the point cloud spacing for adjacent laser points, extract the laser points with a spacing greater than the breakpoint determination threshold as pseudo-breakpoints, and combine the relative positions to determine whether each pseudo-breakpoint is a true breakpoint or a false breakpoint; Among them, a true breakpoint represents an interruption in the horizontal distribution of the point cloud caused by the edge of the wall, and a false breakpoint represents an interruption in the horizontal distribution of the point cloud caused by the obstruction of the lidar's field of view; Step 4: Use the point cloud polar coordinate curvature calculation formula to calculate the curvature in the horizontal direction at the location of each laser point, and store the results as a two-dimensional point cloud curvature matrix based on the coordinates of the point cloud in the two-dimensional matrix; Step 5: Determine whether the category of each laser point belongs to a pseudo-breakpoint or a ground point, and combine the two-dimensional point cloud curvature matrix, the corresponding curvature threshold, and the true / false breakpoint judgment to determine whether the category belongs to a planar point or a wall edge point on the wall; Perform breadth-first search clustering on all points based on the judged category results, including the following steps: (1) Select a point from the unclustered points and use it as the first point in the point cloud sequence to be searched; (2) Take out the first point in the point cloud sequence to be searched as the search center and determine its category; (3) Perform a rationality determination on the four points above, below, left, and right of the search center, and add the points determined to be reasonable to the point cloud sequence to be clustered; (4) Take out the first point in the point cloud sequence to be clustered and determine its category. If its category is the same as that of the search center, add it to the search queue. If its category is different from that of the search center, do not perform clustering processing. If all the points in the point cloud sequence to be clustered have been taken out, add the search center to the searched queue; (5) If the sequence to be searched is not an empty set, return to step (2). If the search queue is an empty set, this search process is closed. Form a point cloud class from all the points in the searched queue, set a category number for it, and store the category number in the two-dimensional clustering result matrix; (6) Terminate the search algorithm when all points have been clustered. If there are points that have not been clustered, return to step (1); Set the same number for the points of the same category and store them as a two-dimensional clustering result matrix, so as to realize the recognition of the ground and different walls in the indoor environment.

2. The method according to claim 1, characterized in that: In step 1, the size of the two-dimensional matrix is specifically determined based on the following formula: N = Θ / dθ H = Φ / dφ In the formula, N and H are the sizes of the horizontal and vertical dimensions of the two-dimensional matrix respectively, Θ and dθ are the horizontal viewing angle and horizontal angular resolution of the lidar respectively, and Φ and dф are the vertical viewing angle and vertical resolution of the lidar respectively; The coordinates of each point cloud in the two-dimensional matrix are specifically determined based on the following formula: Matx = θ / dθ Maty = φ / dφ Where Matx is the abscissa of a point in the two-dimensional matrix, Maty is the ordinate of a point in the two-dimensional matrix, θ is the horizontal angle of a point relative to the lidar, and ф is the pitch angle of a point relative to the lidar. The specific calculation formulas are as follows: θ = atan2(y, x) Where x, y, and z respectively correspond to the three-dimensional coordinates of the laser point relative to the lidar.

3. The method according to claim 2, characterized in that: The elevation angle α of the i-th point in Step 2 i and the elevation angle difference Δη between it and the adjacent point directly above i are obtained based on the following formula: Δα i = α i+1 -α i Where the subscripts i and i + 1 respectively represent the i-th point and the (i + 1)-th point adjacent vertically above it; For α i and Δα i Points that are both less than the corresponding thresholds are marked as ground points and stored as a two-dimensional matrix.

4. The method according to claim 1, characterized in that: The determination of true and false breakpoints in step three specifically includes: First, distinguish between horizontal breakpoints and vertical breakpoints. If there are multiple consecutive pseudo-breakpoints in the horizontal direction, it is determined as a horizontal breakpoint; if it is not a horizontal breakpoint and the distance between two adjacent points in the horizontal direction is greater than the distance threshold, it is determined as a vertical breakpoint; Then determine the true and false of the breakpoint. All horizontal breakpoints are determined as true breakpoints. Among the vertical breakpoints, those within a specific range from the lidar are determined as true breakpoints, and those outside the range are determined as false breakpoints.

5. The method according to claim 3, characterized in that: In step four, to calculate the curvature of each point in the horizontal direction, first based on the polar radius ρ(θ, ф) in the polar coordinate system, use the difference quotient to calculate its discretized first derivative ρ’(θ, ф) and second derivative ρ”(θ, ф): Δθ = mdθ Where ρ(θ + Δθ, ф) and ρ(θ - Δθ, ф) are calculated using the polar radii of the adjacent n points: The above parameters m and n are both selected as appropriate integer values according to actual needs; Then perform the following calculations to first obtain the discrete curvature calculation result κ(θ, ф), and then use the average distance dis(θ, ф) in the horizontal direction to perform distance normalization to obtain the normalized curvature result dis(θ, φ) = (ρ(θ + Δθ, φ) + ρ(θ, φ) + ρ(θ - Δθ, φ)) / 3 6. The method according to claim 1, wherein: In step five, the judgment rules for the categories of each laser point, in addition to the judgment of ground points, pseudo-breakpoints, and true and false breakpoints in steps two to four, also include: Excluding pseudo-breakpoints and ground points, the points with curvature greater than the curvature threshold and the true breakpoints in the determination of true and false breakpoints are determined as wall horizontal edge points; Excluding pseudo-breakpoints and ground points, the points with curvature less than the curvature threshold and the false breakpoints in the determination of true and false breakpoints are determined as wall upper plane points.

7. The method according to claim 6, characterized in that: The rationality judgment in step (3) includes: Determine whether the four points above, below, left, and right of the search center are within the valid coordinate range. If the horizontal coordinate of the point exceeds the upper limit, the horizontal coordinate is assigned 0; if the horizontal coordinate of the point is less than the lower limit, the horizontal coordinate is assigned the total number of horizontal laser beam lines - 1; if the coordinate in the vertical direction exceeds the coordinate range, this point is marked as a pseudo-breakpoint; determine whether the point has been clustered before, and if so, skip this point.

8. The method according to claim 1, characterized in that: When using a single-line lidar to execute this method, a one-dimensional array is used to replace the form of the two-dimensional matrix for point cloud processing and storage in each step, and the ground point marking step in step two is omitted.

Images