Progressive railway ballast bed area extraction method
Through the progressive railway bed area extraction method, voxel filtering, statistical filtering and density clustering are used to solve the problem of low extraction accuracy of middle road bed area in railway point cloud scenarios, and the middle road bed area in large-scale railway point cloud scenarios is realized Accurate extraction and redundant data reduction.
Patent Information
- Application Number
- CN202410926767.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-29
AI Technical Summary
In the railway point cloud scenario, due to the huge scale of the original data and the large number of redundant points, the extraction accuracy of the track bed area is low, especially under the influence of low vegetation and trees in the wild and woodland areas, it is difficult to accurately identify the track bed area.
The progressive railway road bed area extraction method is adopted, and data preprocessing and outlier removal is performed through voxel filtering and statistical filtering algorithms. The data is rotated to make the road bed area extend along the Y axis. The ground point and non-side are separated by fabric filtering and density clustering algorithms. At the ground point, the enclosure box analysis candidate clusters were constructed, and different elevation intervals were analyzed by rasterization to finely extract the track bed area.
It realizes accurate extraction of the track bed area in different railway scenarios, reduces redundant data, improves the extraction accuracy, and provides a feasible solution for the track bed area extraction in railway point cloud scenarios.
Smart Images

Figure CN119049032A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of point cloud scene segmentation technology and railway transportation, and relates to a progressive railway track bed area extraction method. Background Art
[0002] Railway trackbed is an important part of the railway environment, and accurate identification of trackbed areas is crucial for understanding railway scenes. In the current tasks of railway point cloud scene classification and segmentation, the original railway scene dataset obtained by LiDAR scanning is large in scale and contains a large number of redundant points. It is difficult to directly process different scene targets from it, so the trackbed area is often the first object to be identified. In the railway point cloud scene, the trackbed extends along the direction of the railway and has differences in elevation distribution from other objects in the scene. However, due to the wide distribution of the railway network, in some wild and wooded areas, low vegetation and trees close to the trackbed in space bring certain difficulties to the extraction of the trackbed area, which greatly affects the accuracy of the trackbed area extraction results. Summary of the invention
[0003] The purpose of the present invention is to provide a progressive roadbed area extraction method, which solves the problem of low roadbed area extraction accuracy in different railway scenes and reduces redundant points in the scene.
[0004] The technical solution adopted by the present invention is a progressive railway track bed area extraction method, which is specifically implemented according to the following steps:
[0005] Step 1, data preprocessing: down-sample and remove outliers from the input data of the railway track bed area through voxel filtering and statistical filtering algorithms, and perform data rotation;
[0006] Step 2: Extract candidate clusters of the railway track bed area: cluster the scene after data preprocessing, and build bounding boxes for the clustered point clusters for analysis to obtain candidate clusters containing the track bed area;
[0007] Step 3: Railway ballast area optimization: The obtained candidate clusters are rasterized, and different elevation intervals of continuous grids are analyzed to obtain refined extraction results of the ballast area.
[0008] The present invention is also characterized in that:
[0009] Step 1 is as follows:
[0010] Step 1.1, first use the voxel-based filtering method to downsample, and then process the downsampled data through the statistical filtering method to remove outliers;
[0011] Step 1.2, rotate the data so that the extension direction of the track bed area is distributed along the Y axis.
[0012] Step 1.1 is as follows:
[0013] Step 1.1.1, divide the initial point cloud of the railway track bed area into sections and construct a three-dimensional cubic unit grid with equal side lengths. Next, calculate all the points falling into each cubic grid to obtain the centroid positions of these points. Finally, the centroid points calculated for each grid are used to replace all point cloud data points distributed in the grid.
[0014] Step 1.1.2, perform statistical analysis on the neighborhood of each point, calculate the average distance between it and the adjacent points, assume that the calculated average distance result obeys Gaussian distribution, and establish a standard range based on the calculated mean and standard deviation. When the average distance of a point exceeds this standard range, the point is defined as an outlier and removed from the data. Traverse all points and remove outliers.
[0015] Step 2 is as follows:
[0016] Step 2.1, using cloth filtering algorithm to separate ground points and non-ground points in the pre-processed scene, and identifying and extracting non-ground point clouds;
[0017] Step 2.2, clustering the acquired non-ground point cloud data using a density-based clustering algorithm to cluster the non-ground point cloud in the scene into multiple point clusters;
[0018] Step 2.3, construct bounding boxes for each point cluster after clustering, and analyze the side lengths of different bounding boxes to identify candidate clusters containing the roadbed area;
[0019] Step 2.1 is as follows:
[0020] Step 2.1.1, invert the point cloud data preprocessed in step 1, define the mesh edge length and construct the initial cloth mesh;
[0021] Step 2.1.2, project the point cloud and grid particles to the same horizontal plane, find the nearest neighboring point of each particle and record the corresponding elevation value IHV;
[0022] Step 2.1.3, for movable cloth mesh particles, calculate their displacement according to the influence of gravity and compare it with the IHV of its current nearest neighbor. When the height of the particle is less than or equal to IHV, set the particle height to IHV and mark it as an immovable point;
[0023] Step 2.1.4, calculate the displacement of each grid particle caused by the internal driving;
[0024] Step 2.1.5, repeat steps 2.1.3 to 2.1.4, and stop the iteration when the iteration round reaches the set threshold or the maximum elevation change of all particles does not exceed 0.005 meters;
[0025] In step 2.1.6, the height difference between the point cloud and the grid particles is obtained, and the ground points are determined as the points where the distance between the point cloud and the particles is less than the set threshold, and the rest of the point cloud is regarded as non-ground points.
[0026] Step 2.2 is as follows:
[0027] Step 2.2.1: First, construct a KD tree for the input non-ground point cloud data and define two key parameters of the algorithm: neighborhood radius ε and minimum number of neighborhood points P. min , ε defines the neighborhood of a point, P min is the minimum number of points in the neighborhood;
[0028] Step 2.2.2: traverse each data point in the data set and calculate the number of points in its ε neighborhood. If the neighborhood of the point contains at least P min points, then mark them as core points;
[0029] Step 2.2.3: For data points that are not core points, if they are within the ε neighborhood of a core point, mark them as boundary points.
[0030] Step 2.2.4, according to the connection relationship between core points and boundary points, merge them into clusters. If two core points are within each other's ε neighborhood, they are classified into the same cluster; if a boundary point and a core point are within each other's ε neighborhood, the boundary point is classified into the cluster to which the core point belongs;
[0031] In step 2.2.5, the remaining points that are not assigned to any cluster are marked as noise points.
[0032] Step 2.3 is as follows:
[0033] Step 2.3.1, process the point clusters obtained after clustering in sequence. For any point cluster k , calculate the maximum and minimum values of the point cluster in the X, Y, and Z directions, and record them as x max 、x min ,y max ,y min 、z max 、z min ;
[0034] Step 2.3.2, determine the eight vertices of the AABB bounding box by obtaining the maximum values in the X, Y, and Z directions, respectively. min ,y min ,zmin ), B(x max ,y min ,z min ), C(x min ,y max ,z min ), D(x max ,y max ,z min ), D(x max ,y max ,z min ), E(x min ,y min ,z max ), F(x max ,y min ,z max ), G(x min ,y max ,z max ), H(x max ,y max ,z max );
[0035] Step 2.3.3, based on the vertex information of the bounding box, the six face equations of the AABB bounding box can be obtained by using the point-by-point plane equation: ABC:Dz-z min =0, CDGH:yy max =0, BDFH:xx max =0, ABEF:yy min =0, ACGE:xx min =0, EFGH:zz max =0;
[0036] Step 2.3.4, determine the AABB bounding box according to step 2.2.2 and step 2.2.3;
[0037] Step 2.3.5, get the length of each side of the bounding box, and get the distance of the length, width and height of the bounding box AC, AB and AE respectively from the vertex coordinates, that is, for any point cluster k The corresponding length, width and height distances are shown in formula (1).
[0038]
[0039] In formula (1), W k is the width of the bounding box, L k is the length of the bounding box and H k is the height of the bounding box;
[0040] Step 2.3.6: When the distribution of the side length of the bounding box satisfies the following formula (2), the point cluster is confirmed as a candidate cluster;
[0041]
[0042] In formula (2), L k and W k are point clusters k Corresponding to the length and width of the bounding box, max represents the operator for taking the maximum value, {L1,L2,...,L n} is the set of all bounding box lengths.
[0043] Step 3 is as follows:
[0044] Step 3.1, segment the candidate cluster data along the Y-axis direction, the basis of segmentation is determined according to formula (3), and in the subsequent processing steps, the segmented data is processed segment by segment;
[0045]
[0046] In formula (3), p i_y represents the y coordinate of any point p in the candidate cluster, y min Indicates the minimum value of the y coordinate, step is the step length of the segment, P n (n=1,2,3...,n) indicates that the point belongs to the Pth n part.
[0047] Step 3.2, grid division is performed on any candidate cluster of segments after segmentation. The grid division process first creates a blank grid according to the segmentation range and the predefined grid side length. The specific division is performed according to the following formula (4) to determine the row number and column number of the point to complete the segmentation grid division;
[0048]
[0049] In formula (4), row i For point p i The row number, col i is the column number, p i_x 、p i_y Point p i The x- and y-coordinates of the grid, l is the side length of a single grid.
[0050] Step 3.3 analyzes the points inside the divided grids. Starting from the initial grid, the points in three consecutive grids along the X-axis are analyzed in turn. For any grid, the number of points in different elevation areas in the grid is analyzed. According to the span of the point in the Z-axis direction of the ballast cross section, the track point cloud or other ballast area point cloud of the point is determined. The elevation interval for analyzing the grid is given by the following formula (5);
[0051]
[0052] In formula (5), Grid i represents an arbitrary grid, z avg 、z max and z min Represent the average, maximum and minimum values of the z coordinate, G s is the distance between the detection area and the z-axis average value, and dis is the span of the detection area;
[0053] Step 3.4: Count the number of points in different elevation intervals in the continuous grid. For three continuous grids, left , Grid mid , Grid right , when the maximum number of points in different elevation intervals within the grid satisfies the following formula (6), these three grids are marked;
[0054]
[0055] In formula (6), Grid left .num, Grid right .num, Grid mid .num is the maximum number of points in different intervals of the grid, where maxDense and minDense are the thresholds for judging the maximum number of points in the middle and two sides of the grid respectively;
[0056] Step 3.4, obtain the edge part of the roadbed area, divide the entire roadbed area, and calculate the maximum and minimum values of all the points in the marked grid in the X and Y directions, that is, x max 、x min ,y max ,y min , where the maximum and minimum values in the X-axis direction represent the left and right edges of the roadbed area, respectively, and the maximum and minimum values in the Y-axis direction represent the starting point and the end point of the roadbed area, respectively. Based on this, the roadbed area is divided in the candidate cluster. When any point p satisfies the following formula (7), the point is regarded as a part of the roadbed area in the segment.
[0057]
[0058] The beneficial effects of the present invention are:
[0059] The progressive railway track bed area extraction method of the present invention first downsamples and denoises the original point cloud, and rotates the track bed to extend along the axis direction; secondly, the ground points in the scene are eliminated, and the non-ground points are clustered into multiple point clusters using a density-based clustering method, and the bounding boxes constructed by the point clusters are analyzed to identify candidate clusters containing the track bed area; finally, based on the analysis of the elevation changes in different areas on the track bed cross section, the data is rasterized on the basis of the candidate cluster segmentation, and the continuous grid is analyzed to identify the information of the edge part of the track bed, and the refined extraction of the track bed is completed by dividing the edge part in the candidate cluster. The present invention can accurately extract the track bed area in large-scale railway point cloud scenes, while effectively reducing the redundant data in the original scene, and provides a feasible solution for the extraction of track bed areas in railway point cloud scenes. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 A method framework diagram for a progressive railway trackbed area extraction method;
[0061] FIG2 is a result diagram of the progressive railway track bed area extraction method filtering out ground points in the scene, wherein FIG2(a) is the input data, and FIG2(b) is the result of the cloth filtering algorithm processing;
[0062] Figure 3 The result of clustering non-ground points in the scene for the progressive railway trackbed area extraction method;
[0063] FIG4 is a result diagram of candidate cluster extraction of the ballast area by the progressive railway ballast area extraction method, wherein FIG4(a) is a top view of the candidate cluster extraction result, and FIG4(b) is a side view of the candidate cluster extraction result;
[0064] Figure 5 The results of the ballast area extraction using the progressive railway ballast area extraction method and the local detail map. DETAILED DESCRIPTION
[0065] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0066] Example 1
[0067] This embodiment provides a method for extracting a progressive roadbed area. Figure 1 As shown, the specific steps are as follows:
[0068] Step 1, data preprocessing: downsample the input data and remove outliers through voxel filtering and statistical filtering algorithms, and perform data rotation;
[0069] Step 1.1, first use the voxel-based filtering method to downsample, and then use the statistical filtering method to process the downsampled data to remove outliers;
[0070] Step 1.1.1, divide the initial point cloud of the railway track bed area into sections and construct a three-dimensional cubic unit grid with equal side lengths. Next, calculate all the points falling into each cubic grid to obtain the centroid positions of these points. Finally, the centroid points calculated for each grid are used to replace all point cloud data points distributed in the grid.
[0071] Step 1.1.2, perform statistical analysis on the neighborhood of each point, calculate the average distance between it and the adjacent points, assume that the calculated average distance result follows a Gaussian distribution, and establish a standard range based on the calculated mean and standard deviation. When the average distance of a point exceeds this standard range, the point is defined as an outlier and removed from the data. Traverse all points and remove outliers.
[0072] Step 1.2, use CloudCompare software to rotate the data so that the extension direction of the track bed area is distributed along the Y axis;
[0073] Step 2: Extract candidate clusters of the railway track bed area: cluster the scene after data preprocessing, and build bounding boxes for the clustered point clusters for analysis to obtain candidate clusters containing the track bed area;
[0074] Step 3: Railway ballast area optimization: The obtained candidate clusters are rasterized, and different elevation intervals of continuous grids are analyzed to obtain refined extraction results of the ballast area.
[0075] Example 2
[0076] This embodiment provides a progressive roadbed area extraction method, which is implemented based on Embodiment 1 in the following steps:
[0077] Step 1, data preprocessing: down-sample and remove outliers from the input data of the railway track bed area through voxel filtering and statistical filtering algorithms, and perform data rotation;
[0078] Step 2, railway track bed area candidate cluster extraction: cluster the scene after filtering out the ground points, and construct bounding boxes for the clustered point clusters to analyze and obtain candidate clusters containing the track bed area;
[0079] Step 2.1, using cloth filtering algorithm to separate ground points and non-ground points in the scene, identify and extract non-ground point clouds;
[0080] Step 2.1.1, invert the point cloud data preprocessed in step 1 as shown in FIG2(a), define the mesh edge length and construct the initial cloth mesh;
[0081] Step 2.1.2, project the point cloud and grid particles to the same horizontal plane, find the nearest neighboring point of each particle and record the corresponding elevation value (IHV);
[0082] Step 2.1.3, for movable cloth mesh particles, calculate their displacement according to the influence of gravity and compare it with the IHV of its current nearest neighbor. When the height of the particle is less than or equal to IHV, set the particle height to IHV and mark it as an immovable point;
[0083] Step 2.1.4, calculate the displacement of each grid particle caused by the internal driving;
[0084] Step 2.1.5, repeat steps 2.1.3 to 2.1.4, and stop the iteration when the number of iterations reaches the set threshold or the maximum elevation change of all particles is small enough;
[0085] In step 2.1.6, the height difference between the point cloud and the grid particles is obtained, and the ground points are determined as the points where the distance between the point cloud and the particles is less than the set threshold. The remaining point clouds are regarded as non-ground points. The processing result of the cloth filter algorithm is shown in Figure 2(b).
[0086] Step 2.2: Cluster the acquired non-ground point cloud data using a density-based clustering algorithm. Figure 3 As shown, the non-ground point cloud in the scene is clustered into multiple point clusters;
[0087] Step 2.2.1: construct a KD tree with the input non-ground point cloud data and define two key parameters of the algorithm: neighborhood radius ε and minimum number of neighborhood points P. min , ε defines the neighborhood of a point, P min is the minimum number of points in the neighborhood;
[0088] Step 2.2.2: traverse each data point in the data set and calculate the number of points in its ε neighborhood. If the neighborhood of the point contains at least P min points, then mark them as core points;
[0089] Step 2.2.3: For data points that are not core points, if they are within the ε neighborhood of a core point, mark them as boundary points.
[0090] Step 2.2.4, according to the connection relationship between core points and boundary points, merge them into clusters. If two core points are within each other's ε neighborhood, they are classified into the same cluster; if a boundary point and a core point are within each other's ε neighborhood, the boundary point is classified into the cluster to which the core point belongs;
[0091] In step 2.2.5, the remaining points that are not assigned to any cluster are marked as noise points.
[0092] Step 2.3, construct bounding boxes for each point cluster after clustering, and analyze the side lengths of different bounding boxes to identify candidate clusters containing the roadbed area;
[0093] Step 2.3.1, process the point clusters obtained after clustering in sequence. For any point cluster k , calculate the maximum and minimum values of the point cluster in the X, Y, and Z directions, and record them as x max 、x min ,y max ,y min 、z max 、z min ;
[0094] Step 2.3.2, determine the eight vertices of the AABB bounding box by obtaining the maximum values in the X, Y, and Z directions, respectively. min ,y min ,z min ), B(x max ,y min ,z min ), C(x min ,y max ,z min ), D(x max ,y max ,z min ), D(x max ,y max ,z min ), E(x min ,y min ,z max ), F(x max ,y min ,z max ), G(x min ,y max ,z max ), H(x max ,y max ,z max );
[0095] Step 2.3.3, based on the vertex information of the bounding box, the six face equations of the AABB bounding box can be obtained by using the point-by-point plane equation: ABC:Dz-z min =0, CDGH:yy max =0, BDFH:xx max =0, ABEF:yy min =0, ACGE:xx min =0, EFGH:zz max =0;
[0096] Step 2.3.4, determine the AABB bounding box according to step 2.2.2 and step 2.2.3;
[0097] Step 2.3.5, get the length of each side of the bounding box, and get the distance of the length, width and height of the bounding box AC, AB and AE respectively from the vertex coordinates, that is, for any point cluster k The corresponding length, width and height distances are shown in formula (1):
[0098]
[0099] In formula (1), W k is the width of the bounding box, L k is the length of the bounding box and H k is the height of the bounding box.
[0100] Step 2.3.6: further analyze the calculated side lengths of each bounding box. When the distribution of the side lengths of the bounding boxes satisfies the following formula (2), the point cluster is confirmed as a candidate cluster. The candidate cluster extraction result is as follows: Figure 4(a)-Figure 4(b) As shown;
[0101]
[0102] In formula (2), L k and W k are point clusters k Corresponding to the length and width of the bounding box, max represents the operator for taking the maximum value, {L1,L2,...,L n} is the set of all bounding box lengths.
[0103] Step 3: Railway ballast area optimization: The obtained candidate clusters are rasterized, and different elevation intervals of continuous grids are analyzed to obtain refined extraction results of the ballast area.
[0104] Example 3
[0105] This embodiment provides a method for extracting a progressive roadbed area. Based on Embodiment 1-2, the method is specifically implemented according to the following steps:
[0106] Step 1, data preprocessing: down-sample and remove outliers from the input data of the railway track bed area through voxel filtering and statistical filtering algorithms, and perform data rotation;
[0107] Step 2: Extract candidate clusters of the railway track bed area: cluster the scene after data preprocessing, and build bounding boxes for the clustered point clusters for analysis to obtain candidate clusters containing the track bed area;
[0108] Step 3, railway track bed area optimization: grid the obtained candidate clusters, and analyze the different elevation intervals of the continuous grids to obtain the refined extraction results of the track bed area;
[0109] Step 3.1, segment the candidate cluster data along the Y-axis direction, the basis of segmentation is determined according to formula (3), and in the subsequent processing steps, the segmented data is processed segment by segment;
[0110]
[0111] In formula (3), p i_y represents the y coordinate of any point p in the candidate cluster, y min Indicates the minimum value of the y coordinate, step is the step length of the segment, and the step length of the segment is defined as 10m, P n (n=1,2,3...,n) indicates that the point belongs to the Pth n part;
[0112] Step 3.2, grid division is performed on any candidate cluster of segments after segmentation. The grid division process first creates a blank grid according to the segmentation range and the predefined grid side length. The grid size is defined as 0.2m. The specific division is performed according to the following formula (4) to determine the row and column numbers of the points to complete the segmentation grid division;
[0113]
[0114] In formula (4), row i For point p i The row number, col i is the column number, p i_x 、p i_y Point p i The x- and y-coordinates of the grid, l is the side length of a single grid.
[0115] Step 3.3, analyze the points inside the divided grids. Starting from the initial grid, analyze the points in three consecutive grids along the X-axis in turn. For any grid, analyze the number of points in different elevation areas in the grid. According to the span of the point in the Z-axis direction of the ballast cross section, determine whether the point is a track point cloud or other ballast area point cloud. The point cloud with a certain span (set span) in the Z-axis direction of the ballast cross section is a track point cloud, and the point cloud with a smaller span (less than the set span) in the Z-axis direction is other ballast areas. The following formula (5) gives the elevation interval for analyzing the grid;
[0116]
[0117] In formula (5), Grid i represents an arbitrary grid, zavg 、z max and z min Represent the average, maximum and minimum values of the z coordinate, G s is the distance between the detection area and the z-axis average value, and dis is the span of the detection area.
[0118] Step 3.4: Count the number of points in different elevation intervals in the continuous grid. For three continuous grids, left , Grid mid , Grid right , when the maximum number of points in different elevation intervals within the grid satisfies the following formula (6), these three grids are marked;
[0119]
[0120] In formula (6), Grid left .num, Grid right .num, Grid mid .num is the maximum number of points in different intervals of the grid, where maxDense and minDense are the thresholds for judging the maximum number of points in the middle and two sides of the grid respectively;
[0121] Step 3.5, obtain the edge of the roadbed area, divide the entire roadbed area, and calculate the maximum and minimum values of all the points in the marked grid in the X and Y directions, that is, x max 、x min ,y max ,y min , where the maximum and minimum values in the X-axis direction represent the left and right edges of the roadbed area, respectively, and the maximum and minimum values in the Y-axis direction represent the starting point and the end point of the roadbed area, respectively. Based on this, the roadbed area is divided in the candidate cluster. When any point p satisfies the following formula (7), the point is regarded as a part of the roadbed area in the segment.
[0122]
[0123] The results of the ballast area extraction and local details such as Figure 5 shown.
[0124] Through the above method, it can be known that the progressive roadbed area extraction method of the present invention solves the problem of low accuracy of roadbed area extraction in different railway scenes, reduces redundant points in the scene, accurately extracts the roadbed area in large-scale railway point cloud scenes, and effectively reduces redundant data in the original scene, providing a feasible solution for the extraction of roadbed areas in railway point cloud scenes.
Claims
1. A progressive railway track bed area extraction method, characterized in that: The specific steps are as follows: Step 1, data preprocessing: down-sample and remove outliers from the input data of the railway track bed area through voxel filtering and statistical filtering algorithms, and perform data rotation; Step 2: Extract candidate clusters of the railway track bed area: cluster the scene after data preprocessing, and build bounding boxes for the clustered point clusters for analysis to obtain candidate clusters containing the track bed area; Step 3: Railway ballast area optimization: The obtained candidate clusters are rasterized, and different elevation intervals of continuous grids are analyzed to obtain refined extraction results of the ballast area.
2. The method for extracting a progressive railway track bed area according to claim 1, characterized in that: The step 1 is specifically as follows: Step 1.1, first use the voxel-based filtering method to downsample, and then process the downsampled data through the statistical filtering method to remove outliers; Step 1.2, rotate the data so that the extension direction of the track bed area is distributed along the Y-axis.
3. The method for extracting a progressive railway track bed area according to claim 2, characterized in that: The step 1.1 is specifically as follows: Step 1.1.1, divide the initial point cloud of the railway track bed area into sections and construct a three-dimensional cubic unit grid with equal side lengths. Next, calculate all the points falling into each cubic grid to obtain the centroid positions of these points. Finally, the centroid points calculated for each grid are used to replace all point cloud data points distributed in the grid. Step 1.1.2, perform statistical analysis on the neighborhood of each point, calculate the average distance between it and the adjacent points, assume that the calculated average distance result obeys Gaussian distribution, and establish a standard range based on the calculated mean and standard deviation. When the average distance of a point exceeds this standard range, the point is defined as an outlier and removed from the data. Traverse all points and remove outliers.
4. The method for extracting a progressive railway track bed area according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1, using cloth filtering algorithm to separate ground points and non-ground points in the pre-processed scene, and identifying and extracting non-ground point clouds; Step 2.2, clustering the acquired non-ground point cloud data using a density-based clustering algorithm to cluster the non-ground point cloud in the scene into multiple point clusters; In step 2.3, a bounding box is constructed for each point cluster after clustering, and the side lengths of different bounding boxes are analyzed to identify the candidate cluster containing the trackbed area.
5. The method for extracting a progressive railway track bed area according to claim 4, characterized in that: The step 2.1 is specifically as follows: Step 2.1.1, invert the point cloud data preprocessed in step 1, define the mesh edge length and construct the initial cloth mesh; Step 2.1.2, project the point cloud and grid particles to the same horizontal plane, find the nearest neighboring point of each particle and record the corresponding elevation value IHV; Step 2.1.3, for movable cloth mesh particles, calculate their displacement according to the influence of gravity and compare it with the IHV of its current nearest neighbor. When the height of the particle is less than or equal to IHV, set the particle height to IHV and mark it as an immovable point; Step 2.1.4, calculate the displacement of each grid particle caused by the internal driving; Step 2.1.5, repeat steps 2.1.3 to 2.1.4, and stop the iteration when the iteration round reaches the set threshold or the maximum elevation change of all particles does not exceed 0.005 meters; In step 2.1.6, the height difference between the point cloud and the grid particles is obtained, and the ground points are determined as the points where the distance between the point cloud and the particles is less than the set threshold, and the rest of the point cloud is regarded as non-ground points.
6. The method for extracting a progressive railway track bed area according to claim 4, characterized in that: The step 2.2 is specifically as follows: Step 2.2.1: First, construct a KD tree for the input non-ground point cloud data and define two key parameters of the algorithm: neighborhood radius ε and minimum number of neighborhood points P. min , ε defines the neighborhood of a point, P min is the minimum number of points in the neighborhood; Step 2.2.2: traverse each data point in the data set and calculate the number of points in its ε neighborhood. If the neighborhood of the point contains at least P min points, then mark them as core points; Step 2.2.3: For data points that are not core points, if they are within the ε neighborhood of a core point, mark them as boundary points; Step 2.2.4, according to the connection relationship between core points and boundary points, merge them into clusters. If two core points are within each other's ε neighborhood, they are classified into the same cluster; if a boundary point and a core point are within each other's ε neighborhood, the boundary point is classified into the cluster to which the core point belongs; In step 2.2.5, the remaining points that are not assigned to any cluster are marked as noise points.
7. The method for extracting a progressive railway track bed area according to claim 4, characterized in that: The step 2.3 is specifically as follows: Step 2.3.1, process the point clusters obtained after clustering in sequence. For any point cluster k , calculate the maximum and minimum values of the point cluster in the X, Y, and Z directions, and record them as x max 、x min ,y max ,y min 、z max 、z min ; Step 2.3.2, determine the eight vertices of the AABB bounding box by obtaining the maximum values in the X, Y, and Z directions, respectively. min ,y min ,z min ), B(x max ,y min ,z min ), C(x min ,y max ,z min ), D(x max ,y max ,z min ), D(x max ,y max ,z min ), E(x min ,y min ,z max ), F(x max ,y min ,z max ), G(x min ,y max ,z max ), H(x max ,y max ,z max ); Step 2.3.3, according to the vertex information of the bounding box, use the point-by-point plane equation to get the six face equations of the AABB bounding box, which are ABCD:zz min =0, CDGH:yy max =0, BDFH:xx max =0, ABEF:yy min =0, ACGE:xx min =0, EFGH:zz max =0; Step 2.3.4, determine the AABB bounding box according to step 2.2.2 and step 2.2.3; Step 2.3.5, get the length of each side of the bounding box, and get the distance of the length, width and height of the bounding box AC, AB and AE respectively from the vertex coordinates, that is, for any point cluster k The corresponding length, width and height distances are shown in formula (1): In formula (1), W k is the width of the bounding box, L k is the length of the bounding box and H k is the height of the bounding box; Step 2.3.6: When the distribution of the side length of the bounding box satisfies the following formula (2), the point cluster is confirmed as a candidate cluster; In formula (2), L k and W k are point clusters k Corresponding to the length and width of the bounding box, max represents the operator for taking the maximum value, {L1,L2,...,L n } is the set of all bounding box lengths.
8. The method for extracting a progressive railway track bed area according to claim 1, characterized in that: The step 3 is specifically as follows: Step 3.1, segment the candidate cluster data along the Y-axis direction, the basis of segmentation is determined according to formula (3), and in the subsequent processing steps, the segmented data is processed segment by segment; In formula (3), p i_y represents the y coordinate of any point p in the candidate cluster, y min Indicates the minimum value of the y coordinate, step is the step length of the segment, P n (n=1,2,3...,n) indicates that the point belongs to the Pth n part; Step 3.2, grid division is performed on any candidate cluster of segments after segmentation. The grid division process first creates a blank grid according to the segmentation range and the predefined grid side length. The specific division is performed according to the following formula (4) to determine the row number and column number of the point to complete the segmentation grid division; In formula (4), row i For point p i The row number, col i is the column number, p i_x 、p i_y Point p i The x- and y-coordinates of the grid, l is the side length of a single grid; Step 3.3 analyzes the points inside the divided grids. Starting from the initial grid, the points in three consecutive grids along the X-axis are analyzed in turn. For any grid, the number of points in different elevation areas in the grid is analyzed. According to the span of the point in the Z-axis direction of the ballast cross section, the track point cloud or other ballast area point cloud of the point is determined. The elevation interval for analyzing the grid is given by the following formula (5); In formula (5), Grid i represents an arbitrary grid, z avg 、z max and z min Represent the average, maximum and minimum values of the z coordinate, G s is the distance between the detection area and the z-axis average value, and dis is the span of the detection area; Step 3.4: Count the number of points in different elevation intervals in the continuous grid. For three continuous grids, left , Grid mid , Grid right , when the maximum number of points in different elevation intervals within the grid satisfies the following formula (6), these three grids are marked; In formula (6), Grid left .num, Grid right .num, Grid mid .num is the maximum number of points in different intervals of the grid, where max density and min density are the thresholds for judging the maximum number of points in the middle and two sides of the grid respectively; Step 3.4, obtain the edge part of the roadbed area, divide the entire roadbed area, and calculate the maximum and minimum values of all the points in the marked grid in the X and Y directions, that is, x max 、x min ,y max ,y min , where the maximum and minimum values in the X-axis direction represent the left and right edges of the roadbed area, respectively, and the maximum and minimum values in the Y-axis direction represent the starting point and the end point of the roadbed area, respectively. Based on this, the roadbed area is divided in the candidate cluster. When any point p satisfies the following formula (7), the point is regarded as a part of the roadbed area in the segment.
