A point cloud-based railway wagon brace bending identification method

Abstract

The present application provides a kind of based on point cloud's railway freight car brace bar bending identification method, comprising the following steps: S1, the point cloud data of the in-bound freight train is collected;S2, by plane fitting algorithm, extract carriage face, determine the range of carriage point cloud bounding box;S3, according to the relative position of freight train brace bar, set the point cloud bounding box range of brace bar relative to the offset of carriage;S4, according to the range of carriage point cloud bounding box and the point cloud bounding box range of brace bar relative to the offset of carriage, point cloud data is extracted.The point cloud based on railway freight car brace bar bending identification method, by obtaining the point cloud data of in-bound freight train, and is extracted, filtering, segmentation etc.Processing, obtain the point cloud data of each brace bar, then project its point cloud, and bend identification is carried out to its projection, to realize the bending condition of automatic discrimination brace bar, avoid the error produced by artificial judgment, improve the judgment identification precision, while saving manpower, improve the identification efficiency.

Description

Technical Field

[0001] This invention relates to the field of railway transportation technology, and in particular to a method for identifying the bending of railway freight car struts based on point clouds. Background Technology

[0002] Open wagons are a type of railway freight car characterized by being uncovered and having four side panels, with two side panels having hinged doors for easy unloading of goods. This type of wagon is mainly used for transporting bulk goods such as coal or ballast that are not afraid of wind and rain. Open wagons are mainly used for transporting bulk goods such as coal, ore, mining construction materials, timber, and steel. They can also be used to transport light-weight machinery. For some large special-purpose open wagons, struts are installed on the wagon to improve the loading capacity of the wagon. They can also be used as automatic unloading machines for grabbing. The abnormality of the struts can directly affect the safety of freight transport.

[0003] Currently, some stations are equipped with intelligent detection and identification systems for train operation status, such as the "Train On-Site Operation Status Monitoring System" with publication number CN216424427U, or the "Intelligent Monitoring and Automatic Identification System for Train Operation in Railway Stations" with publication number CN211943348U. These systems use cameras and other equipment to take pictures and process and analyze the pictures. However, for open wagons with struts, train inspectors can only manually determine whether the struts are faulty by looking at the pictures. This has problems such as large workload, high cost, poor real-time performance, and human error. Summary of the Invention

[0004] In view of this, the present invention proposes a point cloud-based method for identifying the bending of railway freight car struts. By using the 3D structure of point cloud data, the bending of the struts is detected, and the bending degree of each strut is analyzed through data processing, thereby effectively monitoring the usage of the struts.

[0005] The technical solution of this invention is implemented as follows: This invention provides a method for identifying the bending of railway freight car struts based on point clouds, comprising the following steps:

[0006] S1. Collect point cloud data of freight trains entering the station;

[0007] S2. Extract the carriage surface using a plane fitting algorithm to determine the bounding box range of the carriage point cloud;

[0008] S3. Based on the relative position of the freight train struts, set the point cloud bounding box range for the strut offset relative to the carriage;

[0009] S4. Based on the bounding box range of the point cloud of the carriage and the bounding box range of the point cloud of the strut relative to the carriage, extract the point cloud data Cld of the region of interest of the strut. beam ;

[0010] S5, Cld for point cloud data beam Filtering is performed to obtain point cloud data Cld beam_filter ;

[0011] S6, Cld point cloud data beam_filter The data is segmented to obtain the point cloud data corresponding to each strut;

[0012] S7. Perform point cloud dimensionality reduction projection on the point cloud data corresponding to each strut to obtain the projection map of each strut;

[0013] S8. By using the projection diagrams of each strut, the bending of each strut is identified.

[0014] Based on the above technical solutions, preferably, the solution is based on lidar, which is used to collect point cloud data of freight trains.

[0015] Based on the above technical solutions, preferably, in step S5, the point cloud data Cld beam Filtering is performed using a radius neighborhood statistical filtering algorithm to delete neighboring points within the search radius that do not meet the set number of neighboring points.

[0016] Based on the above technical solution, preferably, in step S6, the point cloud data Cld beam_filter The segmentation process includes the following steps:

[0017] S61, Cld point cloud data beam_filter As a set of candidate points;

[0018] S62. Set an empty set as the clustering result set;

[0019] S63. Select a random point from the set of candidate points;

[0020] S64. Perform a nearest point search on the random point. If the nearest point is less than the given Euclidean distance threshold, then include the nearest point in the clustering result set.

[0021] S65. Repeat step S64 until the nearest point of the searched random point is greater than the given Euclidean distance threshold, then stop the search for the nearest point of the random point.

[0022] S66. Remove the points in the clustering result set from the candidate point set and proceed to step S62 until the candidate point set becomes an empty set.

[0023] Based on the above technical solutions, preferably, in step S7, the projection diagram is obtained by projecting the point cloud data corresponding to the strut onto the length direction of the carriage.

[0024] More preferably, with the length direction of the carriage as the z-axis, the width direction as the y-axis, and the height direction as the x-axis, step S7 includes the following sub-steps:

[0025] S71. Set the spacing resolution of each point to resolution;

[0026] S72. Obtain the boundaries of the x, y, and z coordinates of the corresponding strut in the point cloud data, which are (Xmin, Xmax), (Ymin, Ymax), and (Zmin, Zmax), respectively.

[0027] S73. Traverse the point set on the point cloud. For a point P(x0, y0, z0), its coordinates on the image are x = (x0 - Xmin) / resolution, y = (y0 - Ymin) / resolution, and the pixel value is p = (z0 - Zmin). Obtain the projection image of the corresponding strut.

[0028] Based on the above technical solution, preferably, step S8 further includes extracting the connected region of the corresponding strut projection diagram and calculating the bending value of the connected region, and determining whether the strut is bent by the bending value.

[0029] More preferably, the formula for calculating the curvature value Curve is Curve = S area / S min_rect S area S represents the area of ​​the connected region. min_rect This represents the area of ​​the smallest enclosing rectangle.

[0030] Furthermore, a bending threshold is set. threshold When Curve <Curve threshold At that time, it was determined that the corresponding support rod was bent.

[0031] Based on the above technical solutions, the preferred approach is to store the data and recognition results during the recognition process in a database.

[0032] The point cloud-based railway freight car strut bending recognition method of the present invention has the following advantages over the prior art:

[0033] Beneficial effects:

[0034] (1) By acquiring point cloud data of freight trains entering the station and performing extraction, filtering, segmentation and other processing, point cloud data of each support rod is obtained. Then, the point cloud is projected and the projection is bent to achieve automatic judgment of the bending of the support rod. No manual image observation and comparison is required, and errors caused by manual judgment are avoided, which improves the judgment and recognition accuracy, saves manpower and improves recognition efficiency.

[0035] (2) After extracting the strut point cloud data, filter and delete the nearest neighbor points within the search radius that do not meet the set number of points, remove outliers, eliminate the influence of outliers on recognition, improve recognition accuracy, and prepare for subsequent processing.

[0036] (3) The point cloud data (Cld) is extracted by using the bounding box range of the point cloud of the carriage and the bounding box range of the point cloud of the strut offset relative to the carriage. beam This can greatly reduce the number of point clouds to process, eliminate interfering data, and facilitate the next step of data processing;

[0037] (4) Transfer point cloud data to Cld beam_filter By segmenting the data, point cloud data corresponding to each strut can be obtained. This allows for the individual bending recognition of each strut in subsequent steps, further improving the recognition accuracy. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram illustrating the steps of the point cloud-based railway freight car strut bending recognition method of the present invention;

[0040] Figure 2-4 This is a schematic diagram of 3D point cloud data collected from a freight train for the point cloud-based railway freight car strut bending recognition method of the present invention.

[0041] Figure 5 This is a schematic diagram of point cloud data extraction of the region of interest (ROI) of the railway freight car strut bending recognition method based on point cloud in this invention.

[0042] Figure 6 This is a schematic diagram of the region of interest filtering for the strut bending recognition method based on point cloud of the present invention.

[0043] Figure 7-9 This is a schematic diagram of the projection of three crossbeams for the point cloud-based railway freight car strut bending recognition method of the present invention;

[0044] Figure 10 This is a schematic diagram of the circumscribed rectangle of the three beam projections for the point cloud-based railway freight car strut bending recognition method of the present invention. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0046] like Figure 1-10 As shown, the point cloud-based railway freight car strut bending recognition method of the present invention is based on LiDAR. The LiDAR is installed at a fixed position. A gantry-shaped device can be installed on the railway, the width of which is set according to the line conditions. In a specific embodiment, its installation height is selected as 10.5 meters. The LiDAR is installed on the gantry and is located directly above the freight car when the freight car enters the station, with a height difference of about 6 meters from the freight car. At least one LiDAR is configured. During the freight car's entry into the station, the LiDAR can collect point cloud data of the freight train and save the obtained data to the data warehouse for data processing. This method includes steps S1-S8.

[0047] Step S1: Collect point cloud data of freight trains entering the station.

[0048] This step is achieved through a point cloud recognition system. In a specific embodiment, the obtained 3D point cloud data is as follows: Figure 2-4 As shown in the picture, the entire cargo box of the truck and the three support rods inside the cargo box can be seen.

[0049] Step S2: Extract the carriage surface using a plane fitting algorithm to determine the bounding box range of the carriage point cloud.

[0050] Using the length of the carriage as the z-axis, the width as the y-axis, and the height as the x-axis, the carriage surface is extracted using a planar fitting algorithm. Based on the carriage surface, the point cloud region containing the carriage can be obtained. This operation can locate the position of the carriage within the point cloud. Let the bounding box range of the point cloud be S. train (x1, x2), (y1, y2), (z1, z2), in a specific embodiment, the measured range is: X(x1, x2) = (0, 3.74169m), Y(y1, y2) = (-1.58066m, 1.56329m), Z(z1, z2) = (0m, 11.9812m).

[0051] Step S3: Based on the relative position of the freight train struts, set the point cloud bounding box range for the strut offset relative to the carriage.

[0052] Based on the relative position of the truck across the beam, set the point cloud bounding box range S of the corresponding offset of the strut. beam: (x1-offset, x2-offset), (y1-offset, y2-offset), (z1-offset, z2-offset), where offset is the offset of the strut relative to the carriage. In a specific embodiment, the range of the offset X-offset value is (2m, 0m), the range of the Y-offset value is (0.35m, -0.35m), and the range of the Z-offset value is (1m, -2m). It should be noted that in the point cloud bounding box range of setting the offset of the strut relative to the carriage, the point cloud bounding box range of the offset is not the point cloud bounding box range of the strut, but the point cloud bounding box range remaining after removing the point cloud bounding box range of the carriage. Because the point cloud bounding box range of the strut cannot be directly located, it can only be determined by the fixed offset of the strut within the carriage.

[0053] Step S4: Based on the bounding box range of the carriage point cloud and the bounding box range of the strut's offset relative to the carriage, extract the point cloud data Cld of the region of interest for the strut. beam .

[0054] Through S train -S beam The point cloud data Cld of the region of interest of the strut is obtained. beam This determines the bounding box range of the point cloud of the strut, as in one embodiment, such as Figure 5 As shown, the left side is the point cloud data of the carriage before extraction, and the right side is the point cloud data (Cld) of the region of interest for the strut extracted. beam This operation can greatly reduce the number of point clouds to be processed, eliminate interfering data, and facilitate the next step of data processing.

[0055] Step S5: Cld point cloud data beam Filtering is performed to obtain point cloud data Cld beam_filter .

[0056] Cld for point cloud data beam Filtering is performed using a radius neighborhood statistical filtering algorithm, which deletes neighboring points within the search radius that do not meet the set number of nearest neighbors. This can be used to remove outliers. In a specific embodiment, the search radius is set to 0.05m, and the number of nearest neighbors for the query point is 6, resulting in the filtered strut point cloud data Cld. beam_filter ,like Figure 6 As shown, the left side is the point cloud data Cld before filtering. beam The right side shows the filtered point cloud data of the strut (Cld). beam_filter .

[0057] Step S6: Configure the point cloud data using Cld beam_filterThe data is segmented to obtain the point cloud data corresponding to each strut.

[0058] In one embodiment, the railway freight car has three support struts, which hold the cloud data Cld. beam_filter After segmentation, point cloud data corresponding to the three struts can be obtained.

[0059] Among them, for point cloud data Cld beam_filter The segmentation is achieved by executing a distance-based clustering algorithm in Euclidean space, specifically including steps S61-S66.

[0060] Step S61: Configure the point cloud data using Cld beam_filter As a set of candidate points.

[0061] Create an empty set and set the point cloud data using Cld. beam_filter All points are placed into this empty set to form the set of candidate points.

[0062] Step S62: Set an empty set as the clustering result set.

[0063] The clustering result set is used to store the point cloud data of the same strut and label it, thereby separating the points in the point cloud data by each strut.

[0064] Step S63: Select a random point from the set of candidate points.

[0065] The random point is one of the remaining points in the candidate point set. It should be noted that the candidate point set mentioned here may not be the point cloud data Cld contained in step S61. beam_filter As steps S62-S66 continue, the number of points in the set of candidate points will decrease.

[0066] Step S64: Perform a nearest point search on the random point. If the nearest point is less than the given Euclidean distance threshold, then include the nearest point in the clustering result set.

[0067] For the random points selected in step S63, perform a nearest point search. If the nearest point is less than the given Euclidean distance threshold, then include the nearest point in the corresponding clustering result set. This step can summarize the nearest points of the random points one by one to obtain a complete set of strut points.

[0068] Step S65: Repeat step S64 until the nearest point of the searched random point is greater than the given Euclidean distance threshold, then stop the search for the nearest point of the random point.

[0069] Repeat step S64 until a complete set of strut points is included in the corresponding cluster result set. This determination method is based on whether the nearest point found is greater than or equal to a given Euclidean distance threshold. If it is greater, the search for random points is stopped, and the nearest point is not included in the cluster result set. At this time, the corresponding cluster result set is the set of points containing a complete strut.

[0070] Step S66: Remove the points in the clustering result set from the candidate point set and proceed to step S62 until the candidate point set becomes an empty set.

[0071] If the clustering result set cannot include the nearest point, the points that were included are removed from the candidate point set, and the process proceeds to step S62. A new empty set is then created as the clustering result set, and steps S63-S66 are repeated until the clustering result set is empty. It should be noted that after the last point of the last strut is summarized, there may be no points available for searching. In other words, if the nearest point cannot be found in step S63, step S65 is skipped and the process proceeds to step S66.

[0072] In a specific embodiment, the number of struts is three, and the point cloud data is obtained by using Cld. beam_filter The data was segmented, resulting in three clustering sets corresponding to the three struts.

[0073] Step S7: Perform point cloud dimensionality reduction projection on the point cloud data corresponding to each strut to obtain the projection map of each strut.

[0074] In a specific embodiment, the three clustering result sets obtained in step S6 are used as the point cloud data of the three struts, and are respectively denoted as Cld. beam_seg1 Cld beam_seg2 Cld beam_seg3 and respectively for Cld beam_seg1 Cld beam_seg2 Cld beam_seg3 Perform projection analysis.

[0075] As a preferred embodiment, since the projection effect of the strut is most significant in the direction perpendicular to the length of the carriage, i.e. on the plane where the x and y axes are located, the projection images of the three struts are obtained by using the x-0-y plane projection.

[0076] The specific projection method includes steps S71-S73.

[0077] Step S71: Set the spacing resolution of each point to resolution.

[0078] Step S72: Obtain the boundaries of the x, y, and z coordinates of the corresponding strut in the point cloud data, which are (Xmin, Xmax), (Ymin, Ymax), and (Zmin, Zmax), respectively.

[0079] Step S73: Traverse the point set on the point cloud. For a point P(x0, y0, z0), its coordinates on the image are x = (x0 - Xmin) / resolution, y = (y0 - Ymin) / resolution, and the pixel value is p = (z0 - Zmin). Obtain the projection image of the corresponding strut.

[0080] like Figure 7-9 As shown, through steps S71-S73, Cld is processed. beam_seg1 Cld beam_seg2 Cld beam_seg3 The projections were performed sequentially. The left side of the image shows the point cloud data before projection, and the right side shows the projection image, resulting in the projection images of the three struts. The state of the struts on this side can be clearly seen from the image.

[0081] Step S8: Identify the bending of each support rod by using the projection diagram of each support rod.

[0082] Before identifying the bending of each strut, it is necessary to extract the connected regions of the corresponding strut projection map and calculate the bending value of the connected regions. The bending value is used to determine whether the strut is bent.

[0083] Specifically, the formula for calculating the curvature value Curve is Curve = S area / S min_rect S area S represents the area of ​​the connected region. min_rect The area of ​​the smallest enclosing rectangle, such as Figure 10 As shown, the left side is a projection of the three struts, and the right side is the projection of the minimum circumscribed rectangle outside the projection of the three struts, where S is the smallest possible value under the most ideal conditions. min_rect =S area When the degree of bending increases, S min_rect >S area In other words, the greater the bending amplitude of the strut, the smaller the Curve value.

[0084] When performing bend detection, a bend threshold (Curve) is set. threshold When Curve <Curve threshold When the corresponding strut is bent, it is determined that the strut is bent. That is, a bending range is set; when this range is exceeded, the strut is identified as bent. In a specific embodiment, Curve... threshold =0.7.

[0085] In this embodiment, the data of the identified bending of the cross beam of the freight train is stored in a database for easy access and maintenance by the client. The main stored data includes: segmented cross beam point cloud, images of the detected bending, and other data, which can be displayed by the client for staff to make judgments.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying the bending of railway freight car struts based on point clouds, characterized in that, Includes the following steps: S1. Collect point cloud data of freight trains entering the station; S2. Extract the carriage surface using a plane fitting algorithm to determine the bounding box range of the carriage point cloud; S3. Based on the relative position of the freight train struts, set the point cloud bounding box range for the strut offset relative to the carriage; S4. Based on the bounding box range of the point cloud of the carriage and the bounding box range of the point cloud of the strut relative to the carriage, extract the point cloud data to obtain the point cloud data of the region of interest of the strut. ; S5, Point Cloud Data Filtering is performed to obtain point cloud data. ; S6, Transfer point cloud data The data is segmented to obtain the point cloud data corresponding to each strut; Point cloud data The segmentation process includes the following steps: S61, Transfer point cloud data As a set of candidate points; S62. Set an empty set as the clustering result set; S63. Select a random point from the set of candidate points; S64. Perform a nearest point search on the random point. If the nearest point is less than the given Euclidean distance threshold, then include the nearest point in the clustering result set. S65. Repeat step S64 until the nearest point of the searched random point is greater than the given Euclidean distance threshold, then stop the search for the nearest point of the random point. S66. Remove the points in the clustering result set from the candidate point set and proceed to step S62 until the candidate point set becomes an empty set. S7. Perform point cloud dimensionality reduction projection on the point cloud data corresponding to each strut to obtain the projection map of each strut; S8. Extract the connected regions of the corresponding support rod projections from the projection diagrams of each support rod, and calculate the bending value of the connected region. Determine whether the support rod is bent based on the bending value. The calculation formula is = / ,in The area of ​​the connected region. Set a bending threshold for the area of ​​the minimum enclosing rectangle. ,when < At that time, it was determined that the corresponding support rod was bent.

2. The point cloud-based railway freight car strut bending recognition method as described in claim 1, characterized in that, This is achieved using lidar, which is used to collect point cloud data from freight trains.

3. The point cloud-based railway freight car strut bending recognition method as described in claim 1, characterized in that, In step S5, the point cloud data Filtering is performed using a radius neighborhood statistical filtering algorithm to delete neighboring points within the search radius that do not meet the set number of neighboring points.

4. The point cloud-based railway freight car strut bending recognition method as described in claim 1, characterized in that, In step S7, the projection image is obtained by projecting the point cloud data corresponding to the strut onto the length direction of the carriage.

5. The point cloud-based railway freight car strut bending recognition method as described in claim 4, characterized in that, With the length direction of the carriage as the z-axis, the width direction as the y-axis, and the height direction as the x-axis, step S7 includes the following sub-steps: S71. Set the spacing resolution of each point to resolution; S72. Obtain the boundaries of the x, y, and z coordinates of the corresponding strut in the point cloud data, which are (Xmin, Xmax), (Ymin, Ymax), and (Zmin, Zmax), respectively. S73. Traverse the point set on the point cloud. For a point P (x0, y0, z0), its coordinates on the image are x = (x0-Xmin) / resolution, y = (y0-Ymin) / resolution, and the pixel value is p = (z0-Zmin). Obtain the projection image of the corresponding strut.

6. The point cloud-based railway freight car strut bending recognition method as described in claim 1, characterized in that, The data from the recognition process and the recognition results are stored in a database.

Images