Railway clearance detection method and system based on 3D point cloud

Abstract

The invention relates to a railway clearance detection method and system based on 3D point cloud, and the method comprises the steps: obtaining 3D point cloud data through a three-dimensional laser scanner disposed in an area within a platform yellow line, carrying out the preprocessing of the 3D point cloud data, and extracting the point cloud data of a platform edge and a rail; further extracting calculation features including the track spacing, the horizontal distance between the platform and the track and the vertical distance, performing visualization processing to obtain horizontal and vertical limit measurement results between the platform and the track, and giving an early warning when the horizontal and vertical limit measurement results exceed the limit range; efficient format conversion and coordinate calibration of the high-density point cloud are realized by adopting a dynamic partitioning strategy and an RA NSAC algorithm; dynamically calibrating the extension direction of the orbit through a multidirectional growth algorithm and constructing a polynomial fitting model; realizing millimeter-level precision horizontal / vertical limit characteristic calculation based on track curve discretization sampling; and three-dimensional visual analysis is completed by combining algorithm grid reconstruction such as ball rotation and a dynamic coloring early warning mechanism. The traditional manual measurement is broken through, and automatic and high-precision limit detection is realized.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a railway clearance detection method and system based on 3D point cloud. Background Art

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

[0003] The horizontal and vertical clearances between railway platforms and tracks are critical parameters for safe train operation. Traditional manual measurement techniques for these clearances primarily rely on contact measurement methods, such as track gauges, laser rangefinders, and total stations, which manually acquire discrete point data. For example, track gauges measure the distance between the track and the platform through mechanical contact, while total stations use the principle of triangulation to obtain high-precision coordinates.

[0004] However, these traditional measurement methods have a series of problems:

[0005] It needs to be carried out during the train suspension period (such as 0:00-4:00 at night), the time schedule is limited, and the night operation environment is complex, which increases the difficulty of measurement.

[0006] Operators need to enter the track area, which leads to safety hazards such as electric shock and collision.

[0007] A single measurement can only obtain a limited number of sample points (usually 5-10 meters apart), making it difficult to construct a continuous spatial model and easily missing local deformations. There is a significant efficiency bottleneck. Taking a high-speed railway platform as an example, it takes a manual team 4 hours to complete a 500-meter limit measurement, of which 60% of the time is spent on equipment installation and data recording.

[0008] In addition, traditional measurement methods are greatly affected by human factors, and data accuracy and reliability are difficult to guarantee.

[0009] Some existing technologies attempt to replace manual inspection with image recognition. For example, laser sensors can be used to capture 3D point cloud data, which is then processed through a series of algorithms to perform clearance inspection using image reconstruction. However, these methods still require data collection to be performed during train operating hours, and their detection efficiency remains suboptimal. Summary of the Invention

[0010] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a railway clearance detection method and system based on 3D point cloud, which can realize high-precision and automated platform and track clearance analysis and eliminate the need for manual intervention.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] A first aspect of the present invention provides a railway clearance detection method based on 3D point cloud, comprising the following steps:

[0013] Use a 3D laser scanner placed within the yellow line of the platform to acquire and pre-process 3D point cloud data;

[0014] Extract the point cloud data of the platform edge and track from the pre-processed 3D point cloud data;

[0015] From the point cloud data of the platform edge and the track, computational features including track spacing, horizontal distance between the platform and the track, and vertical distance are extracted;

[0016] By performing 3D image reconstruction (visualization) of the platform edge, track point cloud data, and the resulting calculated features, we can obtain the horizontal and vertical clearance measurements between the platform and the track, and issue an early warning when the clearance range is exceeded.

[0017] During preprocessing, the location of the 3D laser scanner on the platform is taken as the origin, and the point cloud data of the square area where the origin is located is cropped to determine the height range of the point cloud corresponding to the origin;

[0018] Based on the platform height threshold, the point cloud data of the area below the platform is extracted to determine the point closest to the origin in the plane where the track is located. The coordinates of this point are the direction vector v growing towards the track side. grow ;

[0019] On the platform horizontal plane along the growth direction vector v grow Select multiple growth starting points in parallel directions, all starting points along v grow Direction of dynamic expansion, and with v grow The included angle is acute, forming a growing array covering the entire length of the track in the cropped point cloud;

[0020] According to the growth array, each growth path v grow The points around the direction extreme point are used to determine the boundary point set, fit the straight line equation of the xy plane, determine the normalized track direction normal vector d, construct the rotation matrix R by aligning it with the standard direction and perform coordinate transformation, and establish a standard coordinate system with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis.

[0021] Furthermore, the point cloud data of the platform edge and track are extracted from the pre-processed 3D point cloud data, specifically:

[0022] Extract the point cloud belonging to the track from the pre-processed 3D point cloud data, determine the point with the minimum distance from the point cloud in the two-dimensional plane to the origin, and judge the relative position relationship between the track and the platform based on the y-coordinate of the minimum distance point, and determine the corresponding growth strategy;

[0023] A platform and track point cloud extraction method based on segmented slicing and adaptive growth grows the platform edge and track point cloud in each slice, and extracts the complete platform edge and track point cloud.

[0024] Furthermore, the point cloud belonging to the track is extracted from the preprocessed 3D point cloud data, and the minimum distance point from the point cloud in the two-dimensional plane to the origin of the area is determined. According to the y-coordinate of the minimum distance point, the relative position relationship between the track and the platform is judged, and the corresponding growth strategy is determined, specifically:

[0025] Using the z-coordinate relative threshold, the point cloud P of the area below the platform is extracted from the pre-processed 3D point cloud data. below , that is, the point cloud to which the track belongs, calculate the minimum distance point from the point cloud in the two-dimensional plane to the origin and its coordinates;

[0026] If the y-coordinate of the minimum distance point is positive, the track is located outside the platform, that is, in the positive direction of the y-axis, and the positive growth strategy is adopted;

[0027] If the y-coordinate of the minimum distance point is negative, the track is located inside the platform, that is, in the negative direction of the y-axis, and the reverse growth strategy is adopted.

[0028] Furthermore, a platform and track point cloud extraction method based on segmented slicing and adaptive growth is used to grow platform and track point clouds within each slice, extracting the complete platform edge and track point cloud, including the following steps:

[0029] Determine the preprocessed 3D point cloud data P cloud In the range of the x-axis direction, the point cloud data is divided into several subsets P by slicing at set intervals along the x-axis. slice,i , each subset P slice,i Corresponding to a slice area;

[0030] Extract each slice P by height and width constraints slice,i The corresponding point cloud P near the platform station,i , use the RANSAC algorithm to fit and filter out the platform plane with the normal vector z component, and search for the platform edge point P with a set step size within the narrow band range of the slice center edge,i , and use this as a seed to grow bidirectionally along the x-axis to generate a complete platform edge point cloud

[0031] Based on the y coordinate position y of the platform edge point edge,i , define the orbital search range outside the edge According to the height constraint and CSF ground filtering algorithm, the point cloud above the track and the ground point cloud are filtered out. edge,i The relative y-coordinate range threshold divides the track area into the platform near side P raill,iand the far side of the platform Two parts;

[0032] Use the RANSAC algorithm to fit the track plane and remove noise to determine the track top surface height z rail,i And extract edge points As a seed, a complete track point cloud is generated through bidirectional growth.

[0033] Furthermore, computational features including track spacing, horizontal distance between the platform and the track, and vertical distance between the platform and the track are extracted from the point cloud data of the platform and the track. The point cloud data is reduced to the xy plane through the projection matrix, and sampling is performed along the x-axis with a set step size. The track edge feature point with the smallest y-coordinate is extracted in each slice, and the track geometry is described by fitting.

[0034] Furthermore, the point cloud data of the platform and track and the obtained calculated features are visualized. Specifically, the point cloud data of the platform edge and track are reconstructed to generate a triangular mesh model. Based on the obtained calculated features, three-dimensional space annotation is performed, and visualization is achieved through rendering.

[0035] A second aspect of the present invention provides a system for implementing the above method, comprising:

[0036] The point cloud data acquisition and pre-processing module is configured to: use a 3D laser scanner arranged within the area within the yellow line of the platform to acquire and pre-process 3D point cloud data;

[0037] The target point cloud segmentation module is configured to: extract the point cloud data of the platform edge and the track from the pre-processed 3D point cloud data;

[0038] The clearance feature analysis module is configured to: extract computational features including track spacing, horizontal distance between platform and track, and vertical distance from platform to track from point cloud data of platform edge and track;

[0039] The image reconstruction and warning module is configured to: perform 3D image reconstruction of the platform edge, track point cloud data, and the calculated features, obtain the horizontal and vertical clearance measurements between the platform and the track, and issue a warning when the clearance range is exceeded;

[0040] During preprocessing, the location of the 3D laser scanner on the platform is taken as the origin, and the point cloud data of the square area where the origin is located is cropped to determine the height range of the point cloud corresponding to the origin;

[0041] Based on the platform height threshold, the point cloud data of the area below the platform is extracted to determine the point closest to the origin in the plane where the track is located. The coordinates of this point are the direction vector v growing towards the track side. grow ;

[0042] On the platform horizontal plane along the growth direction vector v grow Select multiple growth starting points in parallel directions, all starting points along v grow Direction of dynamic expansion, and with v grow The included angle is acute, forming a growing array covering the entire length of the track in the cropped point cloud;

[0043] According to the growth array, each growth path v grow The points around the direction extreme point are used to determine the boundary point set, fit the straight line equation of the xy plane, determine the normalized track direction normal vector d, and establish a standard coordinate system with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis through alignment with the standard direction and coordinate transformation.

[0044] A third aspect of the present invention provides a computer-readable storage medium.

[0045] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the above-mentioned railway clearance detection method based on 3D point cloud.

[0046] A fourth aspect of the present invention provides a computer device.

[0047] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned railway clearance detection method based on 3D point cloud are implemented.

[0048] Compared with the existing technology, one or more of the above technical solutions have the following beneficial effects:

[0049] 1. Considering the limitations of train suspension periods, this solution uses a 3D laser scanner deployed within the platform's yellow line as the origin. After a series of processing, a standard coordinate system is established with the platform's lowest point as the z = 0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis. This standardizes the collected data, allowing subsequent point cloud data processing and related image reconstruction and generation processes to be performed based on this standard coordinate system. This improves the overall versatility of the solution, freeing it from the limitations of train operating time and measurement point layout, and making it adaptable to a wider range of detection scenarios.

[0050] 2. Existing methods for boundary-limited scenarios typically rely on local analysis based on manually selected points or global track line fitting and analysis. These methods suffer from coarse analysis granularity, susceptibility to noise, and accuracy loss due to fitting. This solution, based on a platform and track point cloud extraction method using segmented slicing and adaptive growth, adaptively grows platform and track point clouds within each slice, ultimately extracting a complete platform and track point cloud to support subsequent high-precision boundary feature analysis at any point. Furthermore, thanks to the parallel processing capabilities of the segmented slicing design, this method significantly improves extraction efficiency.

[0051] 3. Compared with traditional contact measurement, which requires operators to enter the track area and poses risks of electric shock and collision, this solution uses a three-dimensional laser scanner to collect 3D point cloud data within the yellow line of the platform. After a series of image processing and image reconstruction, non-contact measurement is achieved, completely eliminating safety hazards. Traditional methods are limited to the period when trains are out of service (mostly at night), while the present invention supports data collection during all time periods and has flexible operating hours. For complex scenarios (such as curved track sections), traditional measurements result in centimeter-level errors due to sparse sampling points. This solution achieves millimeter-level accuracy through high-density point clouds and high-precision target point cloud extraction algorithms. Multi-site registration and curvature correction technology ensure adaptability to a variety of scenarios such as straight lines and curves, and has strong versatility.

[0052] 4. Compared to traditional methods that rely on manual measurement, which is time-consuming, error-prone, and requires only a limited number of measurement points, this solution automates the entire process, significantly improving efficiency and reliability. A segmented slicing algorithm automatically segments the point cloud, polynomial fitting corrects limit features, and surface reconstruction and dynamic early warning generate visualization results, without requiring any human intervention. Compared to manual measurement of a 500-meter platform, which takes four hours and only measures a few fixed points, the data collection process of this invention can be shortened to about an hour, and the entire 500-meter platform range can be fully inspected at selectable intervals. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0054] Figure 1 This is a schematic diagram of a railway clearance detection process based on 3D point cloud provided by one or more embodiments of the present invention;

[0055] Figure 2 is a schematic diagram of a track direction dynamic calibration process provided by one or more embodiments of the present invention;

[0056] Figure 3 is a schematic diagram of 3D point cloud data results collected by a 3D laser scanner provided by one or more embodiments of the present invention;

[0057] Figure 4 is a schematic diagram of target point cloud segmentation and extraction provided by one or more embodiments of the present invention;

[0058] Figure 5 It is a schematic diagram of visualization and early warning of limit feature analysis results provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0060] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0061] Example 1:

[0062] like Figure 1 As shown, the railway clearance detection method based on 3D point cloud includes the following steps:

[0063] S1. Data Acquisition and Automatic Preprocessing. Use a 3D laser scanner to collect high-precision 3D point cloud data (i.e., a collection of 3D coordinates of a large number of points in space) within the platform's yellow line. This data is then preliminarily processed, including format conversion and coordinate alignment, to generate standardized data suitable for subsequent analysis.

[0064] S2. Target point cloud segmentation and extraction. By designing a series of intelligent point cloud processing algorithms, key targets are identified and extracted from the preprocessed 3D point cloud data: the point cloud data of the platform edge and the track, achieving precise separation of the target objects.

[0065] S3. Boundary feature analysis. Determine the boundary feature analysis range and, based on the optional analysis interval, perform distance calculations on the segmented target point cloud (including track spacing, horizontal and vertical distances between platforms and tracks, etc.).

[0066] S4. Visualization and early warning. The processing results of steps S2 and S3 are displayed in an intuitive visual form, and positions that exceed the limit range are automatically detected and early warning is performed to indicate potential safety risks.

[0067] The data collection and preprocessing in step S1 are the foundation of the entire method, ensuring the accuracy and reliability of subsequent analysis. Specifically:

[0068] 1) Use a station-mounted 3D laser scanner to scan the platform and track area at multiple key locations to obtain high-precision point cloud data.

[0069] By obtaining the relative coordinates of the scanning sites, the automatic registration of multi-site scanning data is realized, and finally the complete platform track point cloud data is generated. The coordinate origin defaults to the position of the first scanning site (usually about 2m above the platform plane), such as Figure 3 shown.

[0070] 2) For the high-density point cloud data (LAS format) collected in railway scenes, a dynamic blocking strategy, stream processing mechanism and memory optimization merging strategy were designed to break through the memory limitations of a single machine and safely and efficiently convert the original point cloud data into PCD format (retaining only x, y, and z coordinate information).

[0071] 3) For the limit scenario, this embodiment provides a track direction dynamic calibration method, which combines the platform horizontal plane elevation to determine a standard coordinate system, providing a basis for subsequent track segment extraction and point selection calculation (x direction), platform track relative position determination (y direction), height threshold determination (z direction), etc., so that the entire solution can adapt to different stations and meet the requirements of versatility. Figure 1 , which includes the following steps:

[0072] a) Through straight-through filtering, the point cloud within the square area near the origin (x, y∈[-5m, 5m]) is clipped, and the platform horizontal plane is extracted using the RANSAC algorithm. Its equation is ax+by+cz+d=0. The z component of the normal vector c>0.9 is constrained to ensure the correctness of the extracted platform horizontal plane, and the height range of this part of the point cloud [z min ,z max ].

[0073] b) Establish a dynamic threshold based on the platform height and extract the point cloud of the area below the platform (z∈[z min -5m,z min -0.2m]), this part of the point cloud is the point cloud containing the track, and the point closest to the origin in the two-dimensional plane (xy plane) is determined. The coordinates of this point are used as the direction vector v that grows towards the track side. grow =(v x ,v y ).

[0074] c) Multi-starting point growth strategy: On the platform horizontal plane, along the growth direction vector v grow Select multiple growth starting points in parallel directions, starting point p i =(x start ,y start ) coordinate constraints are x start v x +y start v y >0, that is, with v growThe included angle is acute, forming a growth array that covers the entire length of the track in the cropped point cloud as much as possible.

[0075] d) Iterative point cloud growth algorithm: Each starting growth point along v grow The direction is dynamically expanded, and in each iteration:

[0076] i. Use the straight-through filter to adaptively expand the point cloud near the growth point (x, y extension 0.1m, z range ±0.1m) as the candidate point cloud;

[0077] ii. Select the point along the growth direction v from the candidate point cloud grow The extreme point (the farthest point) is used as the next growth point;

[0078] iii. Determine the continuity of the point cloud in real time and terminate the iteration when there are no points in the search area.

[0079] e) Extract v for each growth path grow The points around the direction extreme point are taken as the boundary point set, and the weighted least squares method is used to fit the straight line equation y=kx+b in the xy plane. The normalized track direction normal vector is calculated based on the straight line equation

[0080] 4) Construct the rotation matrix R by aligning the orbital normal vector d with the standard direction (0, -1, 0) and perform the coordinate transformation P final =R·(P origin -[0,0,z min ] T ), a standard coordinate system is established with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis.

[0081] The target point cloud segmentation and extraction in step S2 uses a series of point cloud processing methods to segment the original point cloud data into platform point clouds, track point clouds, and other point clouds for subsequent feature analysis. Specifically, the following steps are included:

[0082] 1) Track direction adaptive determination and automatic selection of processing strategies. This includes forward growth strategy and reverse growth strategy. Both use the same algorithm. The main difference is that the strategies such as point cloud search direction, boundary point determination and track area selection need to be adapted according to the positive and negative coordinates. The z-coordinate threshold z∈[-0.2,-3] is used to extract the point cloud P below the platform. below , that is, the point cloud to which the track belongs, calculate the minimum distance point and its coordinates from the point cloud in the two-dimensional plane (xy plane) to the origin of the area: in, Represents the Euclidean distance from a point to the origin. Based on the sign of the y-coordinate of the point with the minimum distance, the position relationship of the track relative to the platform is automatically determined:

[0083] a) If y min >0, the track is located outside the platform (positive direction of the y-axis), and the positive growth strategy is adopted;

[0084] b) If y min <0, the track is located inside the platform (negative direction of the y-axis) and a reverse growth strategy is adopted.

[0085] 2) For boundary scenarios, existing methods typically rely on local analysis based on manually selected points or global track line fitting and analysis. These methods suffer from coarse analysis granularity, susceptibility to noise interference, and precision loss caused by fitting. This embodiment proposes a platform and track point cloud extraction method based on segmented slicing and adaptive growth. This method can adaptively grow platform and track point clouds within each slice, and ultimately achieve the extraction of complete platform and track point clouds to support subsequent high-precision boundary feature analysis at any point. At the same time, thanks to the parallel processing capabilities of the segmented slicing design, this method significantly improves extraction efficiency.

[0086] The following takes the positive growth strategy as an example to explain the specific implementation steps in detail:

[0087] a) Segmentation and growth to determine the pre-processed 3D point cloud data P cloud The range in the x-axis direction [x min ,x max ]=[min(P cloud,x ),max(P cloud,x )], and then slice the point cloud along the x-axis at intervals of Δx = 1m to divide it into several subsets P slice,i ={p∈P cloud ∣x min +(i-1)Δx≤p x <x min +iΔx}, each subset corresponds to a slice area.

[0088] For each slice area, the platform and track point cloud will be adaptively grown through steps b), c), and d). By combining the growth results of all slices, the complete platform edge and track point cloud are extracted.

[0089] b) Platform seed point extraction, for each slice P slice,i , extract the point cloud P near the platform by height and width constraints station,i ={p∈P slice,i |-0.3≤p z ≤0.3,-2≤p y ≤2}, the RANSAC algorithm is applied to fit and the platform plane is screened out by the z component of the normal vector, and the narrow band [x i +0.5,x i+0.6] from y = 2 with a step size of Δy = 0.01m to search for the platform edge point P edge,i , and then use this as a seed to grow bidirectionally along the x-axis to generate a complete platform edge point cloud

[0090] c) Rough extraction of track point cloud, based on the y coordinate position y of the platform edge point edge,i , define the orbital search range outside the edge Apply the height constraint z∈[-2,0] and CSF ground filtering to filter out the point cloud above the track and the ground point cloud, and then filter it with y edge,i The relative y coordinate range y∈[y edge,i +0.8,y edge,i +1.6] and y∈[y edge,i +1.6,y edge,i +3] Divide the track area into the platform near side and the far side of the platform Two parts.

[0091] d) Bidirectional growth of platform and track point clouds, applying the RANSAC algorithm to fit the track plane and remove noise points to determine the track top surface height z rail,i And extract edge points As a seed, bidirectional growth is used to generate a complete track point cloud. Point cloud growth uses multi-threaded parallel processing, and the track point cloud sets the search window W rail =[x±0.02,y±0.11,z±0.04] and select the highest point As a new seed, the station point cloud setting window W station =[x±0.02,y±01,z±0.14] and select the maximum value of y and z When there is no point in the window (W∩P slice,i =0) when growth stops.

[0092] 3) Save the segmented target point clouds such as platform point cloud and track point cloud as PCD format files for subsequent 3D image reconstruction.

[0093] The bounded feature analysis in step S3 automatically calculates features of the platform and track point cloud data extracted in step S2, aiming to evaluate whether the spatial relationship between the platform and the track meets safety standards. Specific features calculated include track spacing, horizontal distance between the platform and the track, and vertical distance. To ensure analysis accuracy and adapt to complex scenarios (such as curved segments), this embodiment proposes a refined feature calculation method. Specifically, it includes the following steps:

[0094] 1) The analysis range is determined by the x-axis extreme value of the track point cloud. Using 3D point cloud projection and high-precision polynomial modeling technology, the 3D point cloud is reduced to the xy plane through the projection matrix [[1,0,0],[0,1,0],[0,0,0]], and then sampled along the x-axis with a fine step size of 0.01 meters (optional). The track edge feature point with the smallest y coordinate is extracted in each slice. By fitting a 6th-order polynomial (the equation is The coefficients are in the form of x 6 To the constant terms in descending order), accurately describe the track geometry. This step significantly improves the accuracy of track centerline fitting through high-order polynomial modeling and dense sampling strategies, laying the mathematical foundation for subsequent curvature correction.

[0095] 2) Perform multi-dimensional feature extraction at 0.1m (optional) intervals within the analysis range. i , synchronously extract the three-dimensional slices (thickness 0.03m) of the platform and the two-side tracks, and use a two-stage adaptive algorithm to locate the feature points: in the first stage, the candidate points are determined by extreme value screening, and in the second stage, the neighborhood of the candidate points (such as [p y -0.01,p y ]×[p z ±0.002] area) for secondary precise positioning. The vertical distance calculation introduces anti-noise optimization, through the dynamic slicing in the z direction (range p y ±0.06) and height constraints (preserve z max ±0.02), combined with the first three points averaging strategy, effectively suppresses point cloud noise interference and ensures the stability and repeatability of feature point coordinates.

[0096] 3) Based on the first-order derivative of the track curve Calculate curvature correction coefficient Project the original coordinate difference to the track normal direction. Horizontal distance D h The curvature effect is corrected by multiplying the y difference between the platform and the track midpoint by cosα, and the track spacing D r The vertical distance D is processed by the method of the extreme difference of y values of the two-side tracks. v Directly taking the z-axis difference. This method can reduce the systematic error caused by directly using Euclidean distance calculation and achieve millimeter-level accuracy even on curved segments.

[0097] 4) A multidimensional data structure is used to record the complete characteristic data of each analysis point, including original coordinates, corrected distances, and calculation parameters. The data table is structured with fields such as serial number, location, horizontal / vertical point coordinates, distance value, and track spacing. Output is provided in Excel format, with integrated coordinate visualization and statistical chart generation capabilities. Out-of-limit points can be automatically marked and inspection reports generated.

[0098] The visualization and early warning stage of step S4 mainly displays the analysis results in the form of three-dimensional image reconstruction and issues early warnings for situations that do not meet safety requirements. It specifically includes the following steps:

[0099] 1) Read point cloud files and analysis result files (such as .xlsx format spreadsheet files) to extract data such as mileage position, 3D coordinates, and horizontal / vertical distance limit analysis.

[0100] 2) Use the Alpha Shape algorithm to reconstruct the platform and track point clouds extracted in step S2, and use the multi-scale spherical rotation algorithm to reconstruct other background point clouds, and the resulting triangle mesh model is generated.

[0101] 3) Analyze data at the meter level and implement three-dimensional spatial annotation. This includes constructing red / green lines connecting the horizontal and vertical measurement endpoints, respectively; embedding floating labels at the midpoints of the measurement lines, displaying four-digit distance values and location codes; and setting a threshold response mechanism that automatically switches to a blue warning indicator when the distance exceeds the safety limit.

[0102] 4) A visualization engine is built based on the Open3D framework, and a composite rendering strategy is used to fuse multi-source data. The grid model of the platform track is rendered in bright colors to highlight the main structure, while the background point cloud is colored in slightly darker colors to enhance the depth perception.

[0103] 5) Integrated multimodal interaction features, including support for six-degree-of-freedom scene manipulation (rotation / translation / zoom); point-to-point query functionality, enabling rapid location of measurement point details through spatial indexing; and a layer management panel allowing users to switch the display status of point cloud groups and analysis annotations as needed.

[0104] Compared with traditional contact measurement, which requires operators to enter the track area and poses risks of electric shock and collision, this solution uses a three-dimensional laser scanner to collect 3D point cloud data within the yellow line of the platform (step S1), realizing non-contact measurement and completely eliminating safety hazards. Traditional methods are limited to the period when trains are out of service (mostly at night), while the present invention supports data collection during all time periods and has flexible operating hours. For complex scenarios (such as curved track sections), traditional measurements result in centimeter-level errors due to sparse sampling points. This solution achieves millimeter-level accuracy through high-density point clouds and high-precision target point cloud extraction algorithms (step S2). The multi-site registration of step S1 and the curvature correction technology of step S3 ensure adaptability to a variety of scenarios such as straight lines and curves, and are highly versatile.

[0105] Compared with the traditional method that relies on manual measurement, which is time-consuming, error-prone and has few measurement points, this solution realizes full process automation, significantly improving efficiency and reliability. The segmented slicing algorithm in step S2 automatically segments the point cloud, the polynomial fitting in step S3 corrects the limit features, and step S4 generates visualization results through surface reconstruction and dynamic early warning, without the need for manual intervention throughout the process. Compared with manual measurement of a 500-meter platform that takes 4 hours and only measures a few fixed points, the data collection of the present invention can be shortened to about one hour, and the entire 500-meter platform range can be fully detected according to the optional interval (step S3).

[0106] This case selected a railway platform and track area for clearance monitoring. This area includes approximately 500 meters of track and platform, and requires high-precision detection of horizontal and vertical clearances.

[0107] S1. Data Collection and Preprocessing

[0108] Using a station-based 3D laser scanner, the platform and track areas are scanned at multiple key locations to obtain high-precision point cloud data. By setting the relative coordinates of the scanning stations, automatic registration of multi-station scanning data is achieved, and finally a complete platform and track point cloud data is generated. Figure 3 Using high-density point cloud data (LAS format) collected from a single machine (as shown in the figure), we devised a dynamic blocking strategy, stream processing mechanism, and memory-optimized merging strategy to overcome the memory limitations of a single machine and safely and efficiently convert the raw point cloud data into PCD format. Based on the origin of the station-based scan, we used spatial direct filtering to intercept the core area of the platform. The RANSAC algorithm was then used to iteratively extract the maximum horizontal plane as the platform datum. The platform elevation extremes were calculated as the vertical datum, and the track direction and its normal were calibrated to the x- and y-axes, respectively.

[0109] S2. Target point cloud segmentation and extraction

[0110] By calculating the minimum or maximum value of the point cloud in a certain direction, the relative position relationship between the platform and the track (for example, which side of the track the platform is on) is determined, and different segmentation and processing strategies are selected. According to the relative position relationship between the platform and the track, the corresponding method is called to perform point cloud growth segmentation. This process simulates the growth process of the point cloud and gradually extracts target point clouds such as the platform edge point cloud and the track point cloud. Figure 4 As shown, Figure 4 The blue lines in the figure represent the track point cloud, and the red lines represent the platform edge point cloud. For example, starting from a certain location in the platform point cloud in a slice, the point cloud is gradually grown along the calibrated track direction until a preset condition (such as point cloud density less than a threshold) is met. The growth results of all slices are then combined to separate the platform edge point cloud and the track point cloud. The segmented target point clouds, including the platform edge point cloud and track point cloud, are saved as PCD format files for subsequent analysis and visualization.

[0111] S3. Limit feature analysis

[0112] The starting and ending positions of the subsequent analysis, i.e., the analysis range, are determined based on the minimum and maximum values of the track point cloud in the x-axis direction. In order to more accurately calculate the distance between the platform and the track, a polynomial fit is performed on the track point cloud to correct the horizontal distance. During the fitting process, the track point cloud is projected onto a two-dimensional plane (x-axis and y-axis), and the x-axis is sampled at a certain step size (such as 0.01 meters). The y-value corresponding to each sampling point is calculated to obtain the discrete points of the track curve, and the coefficients of the track curve are then obtained using the polynomial fitting method. Within the determined analysis range, characteristic values such as the horizontal distance, vertical distance, and track spacing between the platform and the track are gradually calculated at set intervals (such as 0.1 meters). The calculated characteristic values such as the horizontal distance, vertical distance, and track spacing are saved as a table file (such as .xlsx format) to facilitate subsequent analysis and visualization.

[0113] S4. Visualization and early warning

[0114] like Figure 5 As shown, after reading the point cloud file and the analysis result file, the Alpha Shape algorithm is used to reconstruct the surface of the platform and track point clouds extracted in step S2 (dark green), and the multi-scale ball rotation algorithm is used to reconstruct other background point clouds (orange). According to the data in the analysis result file, the labels of the horizontal distance (red) and vertical distance (light green), the selected point balls and the connecting lines are drawn in the visualization window. For each analysis position, the selected points used to calculate the horizontal and vertical distances are drawn with balls. According to the values of the horizontal distance and vertical distance, the distance labels are drawn at the corresponding selected point positions, and the corresponding selected points are connected with straight lines to intuitively show the spatial relationship between the platform and the track. The visualization results are shown as follows: Figure 5 As shown in the figure, the lines representing horizontal and vertical distances in the limits are set to different colors. When the horizontal or vertical distance exceeds the preset safety threshold, the color of the line will be set to blue to indicate a warning state.

[0115] Example 2:

[0116] A system for implementing the above method includes:

[0117] The point cloud data acquisition and pre-processing module is configured to: use a 3D laser scanner arranged within the area within the yellow line of the platform to acquire and pre-process 3D point cloud data;

[0118] The target point cloud segmentation module is configured to: extract the point cloud data of the platform edge and the track from the pre-processed 3D point cloud data;

[0119] The clearance feature analysis module is configured to: extract computational features including track spacing, horizontal distance between platform and track, and vertical distance from platform to track from point cloud data of platform edge and track;

[0120] The image reconstruction and warning module is configured to: perform 3D image reconstruction of the platform edge, track point cloud data, and the calculated features, obtain the horizontal and vertical clearance measurements between the platform and the track, and issue a warning when the clearance range is exceeded;

[0121] During preprocessing, the location of the 3D laser scanner on the platform is taken as the origin, and the point cloud data of the square area where the origin is located is cropped to determine the height range of the point cloud corresponding to the origin;

[0122] Based on the platform height threshold, the point cloud data of the area below the platform is extracted to determine the point closest to the origin in the plane where the track is located. The coordinates of this point are the direction vector v growing towards the track side. grow ;

[0123] On the platform horizontal plane along the growth direction vector v grow Select multiple growth starting points in parallel directions, all starting points along v grow Direction of dynamic expansion, and with v grow The included angle is acute, forming a growing array covering the entire length of the track in the cropped point cloud;

[0124] According to the growth array, each growth path v grow The points around the direction extreme point are used to determine the boundary point set, fit the straight line equation of the xy plane, determine the normalized track direction normal vector d, and establish a standard coordinate system with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis through alignment with the standard direction and coordinate transformation.

[0125] Example 3:

[0126] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the railway clearance detection method based on 3D point cloud in the above-mentioned embodiment 1 are implemented.

[0127] Example 4:

[0128] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the railway clearance detection method based on 3D point cloud in the above-mentioned embodiment 1 are implemented.

[0129] The steps or modules involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0130] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. The railway clearance detection method based on 3D point cloud is characterized by: The following steps are involved: Using a 3D laser scanner placed within the yellow line of the platform, 3D point cloud data is acquired and pre-processed to extract point cloud data of the platform edge and track. From the point cloud data of the platform edge and the track, computational features including track spacing, horizontal distance between the platform and the track, and vertical distance are extracted; By reconstructing 3D images of the platform edge, track point cloud data, and the calculated features, we can obtain the horizontal and vertical clearance measurements between the platform and the track, and issue an early warning when the clearance range is exceeded. During the preprocessing, the location of the 3D laser scanner on the platform is taken as the origin, and the point cloud data of the square area where the origin is located is cropped to determine the height range of the point cloud corresponding to the origin; Based on the platform height threshold, the point cloud data of the area below the platform is extracted to determine the point closest to the origin in the plane where the track is located. The coordinates of this point are the direction vector v growing towards the track side. grow ; On the platform horizontal plane along the growth direction vector v grow Select multiple growth starting points in parallel directions, all starting points along v grow Direction of dynamic expansion, and with v grow The included angle is acute, forming a growing array covering the entire length of the track in the cropped point cloud; According to the growth array, each growth path v grow The points around the direction extreme point are used to determine the boundary point set, fit the straight line equation of the xy plane, determine the normalized track direction normal vector d, construct the rotation matrix R by aligning it with the standard direction and perform coordinate transformation, and establish a standard coordinate system with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis.

2. The railway clearance detection method based on 3D point cloud according to claim 1, characterized in that: Extract the point cloud data of the platform edge and track from the pre-processed 3D point cloud data, specifically: Extract the point cloud belonging to the track from the pre-processed 3D point cloud data, determine the point with the minimum distance from the point cloud in the two-dimensional plane to the origin, and judge the relative position relationship between the track and the platform based on the y-coordinate of the minimum distance point, and determine the corresponding growth strategy; A platform and track point cloud extraction method based on segmented slicing and adaptive growth grows the platform edge and track point cloud in each slice, and extracts the complete platform edge and track point cloud.

3. The railway clearance detection method based on 3D point cloud according to claim 2, characterized in that: Extract the point cloud belonging to the track from the preprocessed 3D point cloud data, determine the point with the minimum distance from the point cloud in the two-dimensional plane to the origin, and judge the relative position relationship between the track and the platform based on the y-coordinate of the minimum distance point, and determine the corresponding growth strategy. Specifically: Using the z-coordinate relative threshold, the point cloud P of the area below the platform is extracted from the pre-processed 3D point cloud data. below , that is, the point cloud to which the track belongs, calculate the minimum distance point from the point cloud in the two-dimensional plane to the origin and its coordinates; If the y-coordinate of the minimum distance point is positive, the track is located outside the platform, that is, in the positive direction of the y-axis, and the positive growth strategy is adopted; If the y-coordinate of the minimum distance point is negative, the track is located inside the platform, that is, in the negative direction of the y-axis, and the reverse growth strategy is adopted.

4. The railway clearance detection method based on 3D point cloud according to claim 2, characterized in that: The platform and track point cloud extraction method based on segmented slicing and adaptive growth grows the platform and track point clouds within each slice, extracting the complete platform edge and track point cloud. The method includes the following steps: Determine the preprocessed 3D point cloud data P cloud In the range of the x-axis direction, the point cloud data is divided into several subsets P by slicing at set intervals along the x-axis. slice,i , each subset P slice,i Corresponding to a slice area; Extract each slice P by height and width constraints slice,i The corresponding point cloud P near the platform statio n ,i , use the RANSAC algorithm to fit and filter out the platform plane with the normal vector z component, and search for the platform edge point P with a set step size within the narrow band of the slice center edge,i , and use this as a seed to grow bidirectionally along the x-axis to generate a complete platform edge point cloud 5. The railway clearance detection method based on 3D point cloud according to claim 2, characterized in that: The platform and track point cloud extraction method based on segmented slicing and adaptive growth grows the platform and track point clouds within each slice, extracting the complete platform edge and track point clouds. It also includes the following steps: Based on the y coordinate position y of the platform edge point edge,i , define the orbital search range outside the edge According to the height constraint and CSF ground filtering algorithm, the point cloud above the track and the ground point cloud are filtered out. edge,i The relative y-coordinate range threshold divides the track area into the platform near side and the far side of the platform Two parts; Use the RANSAC algorithm to fit the track plane and remove noise to determine the track top surface height z rail,i And extract edge points As a seed, a complete track point cloud is generated through bidirectional growth.

6. The railway clearance detection method based on 3D point cloud according to claim 1, characterized in that: From the point cloud data of the platform edge and track, computational features including track spacing, horizontal distance between the platform and the track, and vertical distance are extracted. The point cloud data is reduced to the xy plane using a projection matrix. Sampling is performed along the x-axis with a set step size. The track edge feature point with the smallest y-coordinate is extracted in each slice, and the track geometry is described by fitting.

7. The railway clearance detection method based on 3D point cloud according to claim 1, characterized in that: The platform edge and track point cloud data and the obtained calculated features are used to reconstruct a three-dimensional image. Specifically, the point cloud data of the platform edge and track are reconstructed to generate a triangular mesh model. Three-dimensional space annotation is performed based on the obtained calculated features, and the reconstructed three-dimensional image is obtained by rendering.

8. A railway clearance detection method system based on 3D point cloud, characterized by: include: The point cloud data acquisition and pre-processing module is configured to: acquire and pre-process 3D point cloud data using a 3D laser scanner arranged within the area within the yellow line of the platform; The target point cloud segmentation module is configured to: extract the point cloud data of the platform edge and the track from the pre-processed 3D point cloud data; The clearance feature analysis module is configured to: extract computational features including track spacing, horizontal distance between platform and track, and vertical distance from platform to track from point cloud data of platform edge and track; The image reconstruction and warning module is configured to: perform 3D image reconstruction of the platform edge, track point cloud data, and the calculated features, obtain the horizontal and vertical clearance measurements between the platform and the track, and issue a warning when the clearance range is exceeded; During the preprocessing, the location of the 3D laser scanner on the platform is taken as the origin, and the point cloud data of the square area where the origin is located is cropped to determine the height range of the point cloud corresponding to the origin; Based on the platform height threshold, the point cloud data of the area below the platform is extracted to determine the point closest to the origin in the plane where the track is located. The coordinates of this point are the direction vector v growing towards the track side. grow ; On the platform horizontal plane along the growth direction vector v grow Select multiple growth starting points in parallel directions, all starting points along v grow Direction of dynamic expansion, and with v grow The included angle is acute, forming a growing array covering the entire length of the track in the cropped point cloud; According to the growth array, each growth path v grow The points around the direction extreme point are used to determine the boundary point set, fit the straight line equation of the xy plane, determine the normalized track direction normal vector d, and establish a standard coordinate system with the lowest point of the platform as the z=0 plane, the track direction as the x-axis, and the track direction normal vector d as the y-axis through alignment with the standard direction and coordinate transformation.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the railway clearance detection method based on 3D point cloud as described in any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the railway clearance detection method based on 3D point cloud according to any one of claims 1 to 7 are implemented.

Images