Method and device for dynamically measuring reduction value of turnout point rail

Abstract

The invention discloses a dynamic measurement method and device for a turnout sharp point rail reduction value, and the method comprises the steps: carrying out the classification and recognition of turnout point cloud data through a pre-trained multi-scale semantic segmentation model, and outputting the point cloud data of a stock rail and a plurality of turnout structural parts; the turnout structure component comprises a wing rail, a switch rail and a point rail; the multi-scale semantic segmentation model is obtained based on PointNet + + training; on the basis of the least square method, point cloud data of the stock rail and the wing rail are used for fitting to obtain a rail top face curve and a rail head side face curve, and rail vertexes and rail gauge points are obtained; according to the geometric structure relationship between the stock rail and the switch rail, determining a reduction value feature point based on the rail head side curve; and determining a point rail reduction value and fixing the point rail reduction value by utilizing a height difference value between the rail vertex and the reduction value feature point. According to the invention, the accuracy and efficiency of turnout point rail reduction value measurement can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of railway infrastructure, and in particular to a dynamic measurement method and device for a turnout point rail reduction value. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section.

[0003] As a key turning device of the track system, the structural reliability of the turnout directly affects the safety of train operation. The high-frequency impact force and lateral force generated when the train passes through the turnout can easily cause wear and deformation of key components such as the point rail and the heart rail, thereby reducing the service life of the turnout and even causing safety accidents. As the speed and volume of trains continue to increase, higher requirements are placed on the safety and stability of the turnout system. Among them, the reduction value of the turnout point rail and the heart rail (that is, the height difference between the top surface of the point rail and the heart rail relative to the top surface of the base rail) is the core indicator of the wheel-rail contact geometry. It directly affects the wheel load transition characteristics, vehicle vibration and operating speed, and has become a key control indicator for turnout design and maintenance.

[0004] In the field of turnout status detection, manual detection methods are currently mainly relied upon, which is limited by operating experience and visual measurement accuracy, making it difficult to meet the automated detection requirements of the "Standards for Repair of Conventional Railway Lines". Although two-dimensional visual detection technology has been applied in track detection, its planar imaging characteristics are easily affected by changes in lighting, component occlusion, and installation posture, and there is a problem of insufficient robustness in feature extraction in complex turnout scenes. In contrast, three-dimensional point cloud technology can obtain spatial coordinate data containing depth information, which can more accurately describe the three-dimensional geometric features of turnout components. However, existing railway three-dimensional point cloud research is mostly aimed at ordinary track structures (such as fastener and sleeper detection), and turnout scenes have problems such as complex component geometry, sparse turnout point cloud data, and coupling of spatial relationships among multiple components. As a result, traditional point cloud segmentation algorithms are difficult to achieve accurate recognition and efficient processing of turnout structures. Summary of the Invention

[0005] An embodiment of the present invention provides a dynamic measurement method for a turnout point rail drop value, for improving the accuracy and efficiency of the turnout point rail drop value measurement. The method includes:

[0006] Get turnout point cloud data;

[0007] Using a pre-trained multi-scale semantic segmentation model trained based on PointNet++, the turnout point cloud data is classified and identified, and point cloud data of the base rail and multiple turnout structural components are output. The turnout structural components include wing rails, point rails, and point rails.

[0008] Based on the least squares method, the point cloud data of the base rail and wing rail are used to fit the rail top surface curve and rail head side surface curve respectively.

[0009] Based on the rail top curve and rail head side curve, the rail vertex and rail gauge point are obtained;

[0010] According to the geometric structure relationship between the stock rail and the point rail, the reduced value characteristic point is determined based on the side curve of the rail head; the reduced value characteristic point reflects the apex of the point rail or the center rail;

[0011] The height difference between the rail vertex and the reduction value feature point is used to determine the reduction value of the point rail and the fixed center rail.

[0012] An embodiment of the present invention further provides a dynamic measurement device for the turnout point rail reduction value, for improving the accuracy and efficiency of the turnout point rail reduction value measurement, the device comprising:

[0013] Data acquisition module, used to obtain turnout point cloud data;

[0014] A segmentation model processing module is used to classify and identify turnout point cloud data using a pre-trained multi-scale semantic segmentation model, and output point cloud data of the base rail and multiple turnout structural components; the turnout structural components include wing rails, point rails, and point rails; the multi-scale semantic segmentation model is trained based on PointNet++;

[0015] The reduction value calculation module is used to obtain the rail top surface curve and rail head side curve by fitting the point cloud data of the base rail and wing rail based on the least squares method; obtain the rail vertex and track gauge point based on the rail top surface curve and rail head side curve; determine the reduction value feature point based on the rail head side curve according to the geometric structure relationship between the base rail and the point rail; the reduction value feature point reflects the vertex of the point rail or the center rail; and determine the reduction value of the point rail and the fixed center rail using the height difference between the rail vertex and the reduction value feature point.

[0016] An embodiment of the present invention also 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 computer program, the above-mentioned dynamic measurement method of the turnout point rail lowering value is implemented.

[0017] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for dynamically measuring the lowering value of the turnout point rail is implemented.

[0018] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned dynamic measurement method of the turnout point rail lowering value.

[0019] In this embodiment of the present invention, the PointNet++ model is first used to classify and identify turnout point cloud data, outputting point cloud data for the base rail and multiple turnout structural components. This improves the accuracy of component identification in turnout tip rail scenarios. An adaptive least-squares fitting algorithm is then used to extract rail vertices, gauge points, and reduction feature points, enabling high-precision measurement of tip rail and fixed rail reduction values. Experimental verification demonstrates that this method achieves a measurement accuracy of ±0.3mm under dynamic detection conditions, meeting the millimeter-level detection standard for high-speed railway turnouts and improving the accuracy and efficiency of turnout tip rail reduction measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. 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 work. In the drawings:

[0021] Figure 1 Schematic diagram of the flow of a dynamic measurement method for a turnout point rail reduction value according to an embodiment of the present invention;

[0022] Figure 2 A diagram showing a specific example of a method for dynamically measuring a turnout point rail reduction value according to an embodiment of the present invention;

[0023] Figure 3 FIG. 1 is another specific example of a method for dynamically measuring a turnout point rail reduction value in an embodiment of the present invention;

[0024] Figure 4 Schematic diagram of the rail top curve and rail head side curve in an embodiment of the present invention;

[0025] Figure 5 Schematic diagram of the process of calculating the reduction value of the point rail in an embodiment of the present invention;

[0026] Figure 6 Schematic diagram of calculation of reduced value feature points in an embodiment of the present invention;

[0027] Figure 7 Schematic diagram of a dynamic measurement device for the lowering value of the turnout point rail in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0029] There are two situations for the turnout reduction value: one is the reduction value of the point rail relative to the base rail, and the other is the reduction value of the heart rail relative to the wing rail. The reduction value can be obtained by measuring the vertical distance between the feature point and the apex of the base rail and the wing rail. Given that the base rail, point rail and heart rail have significant structural differences, the point rail can be moved to insert or leave the gap between the base rails; the heart rail includes a movable heart rail and a fixed heart rail structure: (1) When it is necessary to open a track in a certain direction, the movable heart rail fork is closely attached to the wing rail in the same opening direction, and at the same time separated from the other wing rail, thereby eliminating the harmful space in the turnout. (2) The wing rails of the fixed heart rail are located on both sides of the heart rail body, and together with the heart rail body, they form a V-shaped structure [25-27]. Therefore, after obtaining the three-dimensional data of the turnout in the track reference coordinate system, due to the complex structural relationship between the point rail and the heart rail, it is necessary to accurately extract the feature points from the scanned data to calculate the reduction value.

[0030] According to existing regulations, when the point rail is lowered by more than 1mm and affects the smoothness of train operation, it must be repaired or replaced. To improve measurement accuracy, it is necessary to accurately identify turnout components and extract feature points.

[0031] To address the technical bottlenecks in the existing technology, the present invention proposes a dynamic measurement method for the point-center rail reduction value based on multi-field-of-view 3D vision, combined with the on-board dynamic detection platform of the China Academy of Railway Sciences. First, high-density turnout point cloud data is acquired through multi-sensor calibration and field of view fusion technology. Second, the PointNet++ point cloud segmentation model is improved, and a turnout partition downsampling strategy is introduced to improve the recognition accuracy of complex scene components. Finally, an adaptive least squares fitting algorithm is used to extract rail top feature points to achieve high-precision measurement of the reduction value.

[0032] Figure 1 FIG. 1 is a flow chart of a method for dynamically measuring the reduction value of a turnout point rail according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0033] Step 101: Obtain turnout point cloud data;

[0034] Step 102: Use a pre-trained multi-scale semantic segmentation model to classify and identify the turnout point cloud data, and output point cloud data of the base rail and multiple turnout structural components; the turnout structural components include wing rails, point rails, and point rails; the multi-scale semantic segmentation model is trained based on PointNet++;

[0035] Step 103: Based on the least squares method, the point cloud data of the base rail and the wing rail are respectively fitted to obtain the rail top surface curve and the rail head side surface curve;

[0036] Step 104: Obtain the rail vertex and the gauge point based on the rail top curve and the rail head side curve;

[0037] Step 105: Determine a reduced value characteristic point based on the rail head side curve according to the geometric structure relationship between the stock rail and the point rail; the reduced value characteristic point reflects the vertex of the point rail or the center rail;

[0038] Step 106: Determine the point rail reduction value and the fixed center rail reduction value by using the height difference between the rail vertex and the reduction value characteristic point.

[0039] The following is a detailed explanation of the dynamic measurement method of the turnout point rail reduction value according to the embodiment of the present invention.

[0040] In step 101, the turnout point cloud data is obtained.

[0041] Using a turnout inspection system based on structured light measurement principles, three-dimensional data collection, processing, and analysis are performed on the track surface and surrounding infrastructure. This 3D turnout data includes not only the stock rail but also turnout components (point rail, point rail, wing rail, and guard rail), sleepers, fasteners, ballast, and other track infrastructure. Therefore, to prevent interference from extraneous structures in turnout reduction measurement, accurate and efficient identification of turnout structural components is required to provide reliable data support for parameter measurement.

[0042] Figure 2 FIG. 1 is a specific example of a method for dynamically measuring the reduction value of a turnout point rail according to an embodiment of the present invention. Figure 2 As shown in the figure, feature points are calculated based on the improved PointNet++ multi-scale semantic segmentation model and the least squares fitting of the turnout profile curve. The specific process is as follows: 3D turnout data is acquired, point cloud categories are calculated based on the multi-scale semantic segmentation model, and turnout structural components are automatically identified. Next, the rail top curve is fitted to calculate the rail vertex, and the rail side curve is fitted to calculate the gauge point. Finally, the contour curve of the area near the gauge point is fitted to calculate the feature points and achieve reduced value measurement.

[0043] In step 102, a pre-trained multi-scale semantic segmentation model is used to classify and identify the turnout point cloud data, and output point cloud data of the base rail and multiple turnout structural components; the turnout structural components include wing rails, point rails, and point rails; the multi-scale semantic segmentation model is trained based on PointNet++.

[0044] In one embodiment, the multi-scale semantic segmentation model may include a sampling layer, a grouping layer, a PointNet layer, a feature propagation layer, and an output layer;

[0045] The sampling layer is used to downsample the input turnout point cloud data using the Farthest Point Sampling (FPS) method.

[0046] The grouping layer is used to construct local area point sets for the downsampled turnout point cloud data based on the K-nearest neighbor algorithm;

[0047] The PointNet layer is used to extract the features of each local area point set;

[0048] The feature propagation layer is used to propagate the features of the local area point set to the original point through interpolation, and output the features that integrate multi-level information;

[0049] The output layer is used to: use the features of multi-level information fusion to perform classification and recognition, and output the turnout structural components corresponding to each point.

[0050] During implementation, hierarchical feature learning and point feature propagation are performed on the turnout point cloud data based on PointNet++.

[0051] (1) Hierarchical feature learning

[0052] Hierarchical feature learning consists of multiple set abstraction layers (SA). At each layer, a set of points is processed and abstracted to generate a new set with fewer elements. The set abstraction layer includes a sampling layer (Sampling), a grouping layer (Grouping), and a PointNet layer. Since the density of the turnout point cloud data in different areas varies significantly, the farthest point sampling strategy FPS is adopted to maintain the point cloud data structure within the divided area in the most balanced way possible. However, this method will destroy the adjacency relationship of the original turnout point cloud data, which will have an adverse effect on the extraction of local geometric structure features and the learning expression of features. Therefore, in one embodiment, the sampling layer is specifically used to:

[0053] The input turnout point cloud data is divided into multiple sub-areas; the sub-areas include the rail head area, the rail waist area and the area below the rail bottom;

[0054] Set the downsampling rate of each sub-region according to the point cloud ratio of the sub-region and the turnout point cloud data;

[0055] According to the downsampling rate of each sub-region, the FPS method is used to downsample each sub-region.

[0056] In this example, a partition-based downsampling method is used, which includes three steps: sub-region division, sampling rate determination, and farthest point sampling. First, the turnout 3D point cloud dataset is denoted as P. Based on the prior information of the rail design structure, the rail top and the gauge point position, the data is divided into three different sub-regions: the rail head region P head , rail waist area P waist and the area below the rail bottom P base , the data volumes are |P head |、|P waist | and |Pbase |; Then, by reducing the amount of data in each region proportionally, we ensure that the proportion of each region in the downsampled dataset is consistent with that in the original dataset. head Downsampling rate r head Expressed as:

[0057]

[0058] Similarly, the downsampling rate r of the rail waist area and the area below the rail bottom can be obtained waist and r base .

[0059] Finally, for each region P i (where i is head, waist or base), and the farthest point sampling method is used to complete the downsampling. Initialize the sampled point set In each sampling step, a point S is selected from the remaining points. i The farthest point is used as the next sampling point. The distance formula is: for point p∈P i and Its to the sampled point set S i The distance is:

[0060] min{s∈S i}||ps|| (2)

[0061] Where s is the set of sampled points S i A point in i It is a set of sampled points, s traverses each point in this set to calculate the distance from the point to be sampled p to the set of sampled points S i The minimum distance. ||·|| represents the Euclidean distance calculation, which means that in the set S i Find s that minimizes ‖ps‖. The actual intention is to traverse the set S i , repeat the above steps until the target amount of data after sampling is reached.

[0062] The grouping layer uses the K-nearest neighbor algorithm to construct local region point sets. The PointNet layer uses the core PointNet network to extract features from local regions. When the input sampling density changes, this layer can learn to combine features of regions of different scales and perform these combinations based on the local point density.

[0063] (2) Point feature propagation

[0064] Downsampling is performed twice in the set abstraction layer. However, semantic segmentation tasks require features from all points. Therefore, a hierarchical propagation strategy based on distance interpolation and cross-layer connections is used to propagate features from subsampled points to the original points. The feature propagation process is as follows: First, the weighted average of the inverse of the k nearest neighbor distances is calculated; then, the interpolated features of the point are connected with the skip-connected point features of the set abstraction layer to form a connected feature; finally, the connected feature is passed through the PointNet unit, and a shared fully connected layer and activation function layer are used to update the feature vector of each point. This process is repeated until the features are propagated to the original point set.

[0065] Figure 3 FIG is another specific example of a method for dynamically measuring the reduction value of a turnout point rail in an embodiment of the present invention, referring to FIG. Figure 3 The data of the base rail, point rail, center rail, and wing rail are obtained through the aforementioned turnout structural component identification results. The reduction value is measured according to different components. First, the rail top surface curve is fitted based on the least squares method. Second, according to the definition of the gauge point, the gauge point located 16 mm below the rail vertex is calculated. Finally, based on the geometric structure relationship of the point and center rail, the reduction value feature point is located in the vicinity of the gauge point. The second-order derivative extreme point is calculated based on the fitting curve of the rail head side. The reduction value feature point is selected using this extreme point and then subtracted from the base rail or wing rail vertex to obtain the point and center rail reduction value measurement result.

[0066] In step 103, based on the least squares method, the point cloud data of the base rail and the wing rail are respectively fitted to obtain the rail top surface curve and the rail head side surface curve.

[0067] During implementation, the basic rail top surface curve and the basic rail head side curve are obtained by fitting the point cloud data of the basic rail, and the wing rail top surface curve and the wing rail head side curve are obtained by fitting the point cloud data of the wing rail.

[0068] In one embodiment, based on the least squares method, fitting the point cloud data of the base rail and the wing rail to obtain the rail top surface curve and the rail head side surface curve may include:

[0069] The point cloud data of the base rail or the wing rail is used to filter and obtain the characteristic points of the rail top surface contour curve;

[0070] Using the standard equation of a circle, an objective function based on the least squares principle is constructed; the objective function includes the parameters to be solved;

[0071] The gradient descent method is used to iteratively adjust the parameters to be solved to minimize the objective function.

[0072] Figure 4Figure 2 is a schematic diagram of the rail top curve and rail head side curve in an embodiment of the present invention. The nonlinear least squares method is used to fit the rail top profile curve. Since the standard rail profile is composed of a series of circular arc curve segments, the rail vertices of the base rail and wing rail can be obtained by fitting the rail top tangent of the profile curve. The arc segment with a rail top radius of 300 mm corresponding to the standard rail is selected for fitting.

[0073] Select the set of the highest N points on the rail top contour curve, and let M be the ordered set of rail vertices on the contour line, that is, M={m i =(x i ,y i )|i=0,1,…,N-1)}, according to the sampling density, N is set to 30. The standard equation of a circle is:

[0074] (xa) 2 +(xb) 2 =R 2 (3)

[0075] In formula (3), the coordinates of the center of the circle are (a, b), R is the radius of the circle, R = 300 mm. The least squares fitting curve L p The goal is to find a and b that minimize the following:

[0076]

[0077] Use the gradient descent method to iteratively adjust a and b to minimize the objective function F(a,b). The gradient is calculated as:

[0078]

[0079] Update the parameters using the learning rate ρ: and When the parameter update change is less than 0.1, the iteration is terminated and the final a and b are obtained, which are the coordinates of the center of the fitting circle.

[0080] In step 104, based on the rail top curve and rail head side curve, the rail vertex and rail gauge point are obtained. Figure 4 , the gauge point is obtained 16 mm below the rail apex.

[0081] For example, according to the profile curve L of the rail top p Calculate the horizontal tangent line L0 of the rail top and translate it downward by 16mm to obtain the straight line L1. Since the rail profile P is a set of discrete points, there may not be an intersection with the straight line L1. By calculating the point P closest to the straight line L1, G =(x k ,y k ) as the track gauge point.

[0082] In step 105, according to the geometrical structural relationship between the stock rail and the point rail, a reduced value characteristic point is determined based on the rail head side curve; the reduced value characteristic point reflects the vertex of the point rail or the center rail.

[0083] In one embodiment, according to the geometrical structural relationship between the stock rail and the point rail, determining the reduction value characteristic point based on the rail head side curve includes:

[0084] The extreme point of the second-order derivative of the rail head side curve is calculated; the extreme point reflects the close contact position between the stock rail and the point rail;

[0085] Calculate the normal line L of the extreme point on the side curve of the rail head I ;

[0086] Determine the distance normal L in the set S I The nearest point R; the set S is the set of base rails and point rails in the turnout point cloud data, or the set of wing rails and center rails in the turnout point cloud data;

[0087] Starting from point R, search backward for the point T with the largest vertical coordinate value in the gauge point setting area, and record it as the reduced value characteristic point.

[0088] During implementation, according to the structural relationship between the point rail and the center rail, the characteristic point with the lowered value is usually within the neighborhood of the gauge point. Therefore, an ordered set S is selected, which contains the point P G And m points before and after: S={(x k-m ,y k-m ),…(x k ,y k ),…(x k+m ,y k+m )}, where x k represents the coordinate value of the k-th point along the rail, and represents the vertical coordinate value of the k-th point.

[0089] The reduction value calculation is divided into the reduction value calculation of the point rail and the fixed center rail. Figure 5 FIG. 1 is a flow chart of calculating the reduction value of the point rail in an embodiment of the present invention. Figure 5 As shown, the calculation process of the point rail reduction value is taken as an example: First, the rail head side curve L is fitted using the above least squares method. S According to the geometric structure of the point rail components, the side profile of the rail head is composed of the base rail and the point rail profile. Therefore, by calculating the curve L S The second-order derivative extreme point Q can locate the close position of the basic rail and the pointed rail; then, calculate the extreme point Q on the curve L S Normal L I , determine the distance normal L in the set S IThe nearest point R, and satisfying |R - Q| < t, where t is the distance threshold, set to 1. If the condition is not met, the fitting parameters need to be updated; finally, starting from point R, search backward for the point T with the largest y value, which is the reduced value feature point, and subtract it from the basic rail vertex to obtain the measured result of the switch reduced value.

[0090] Figure 6 It is a schematic diagram for calculating the reduced value feature point in an embodiment of the present invention. Figure 6 In it, the horizontal coordinate represents the abscissa of the point, and the vertical coordinate represents the ordinate of the point, showing the original data, the fitting curve, the maximum value of the second derivative, the normal direction, the point closest to the normal, and the reduced value feature point.

[0091] In summary, the embodiments of the present invention have the following advantages:

[0092] (1) The embodiments of the present invention propose a dynamic measurement method for the reduced value of the switch nose rail, combining a multi-scale semantic segmentation model and a method of least squares fitting of the profile to achieve the calculation of the reduced value. This method can reduce the labor cost of operation and maintenance and provide a new technical means for railway maintenance.

[0093] (2) The embodiments of the present invention are based on a prior partition downsampling algorithm for the switch structure (region division → proportional sampling → FPS), and a feature propagation mechanism of hierarchical feature learning and cross-layer connection to improve the component recognition accuracy in complex point cloud scenarios.

[0094] (3) The embodiments of the present invention are based on a least squares rail top curve fitting algorithm, a dynamic calculation method for gauge points, and a feature point positioning strategy using the extreme points of the second derivative and geometric constraints to achieve a reduced value measurement with an accuracy of ±0.3 mm, meeting the millimeter-level detection standard for high-speed railways.

[0095] An embodiment of the present invention also provides a dynamic measurement device for the reduced value of the switch nose rail, as described in the following embodiments. Since the principle of solving problems by this device is similar to the dynamic measurement method for the reduced value of the switch nose rail, the implementation of this device can refer to the implementation of the dynamic measurement method for the reduced value of the switch nose rail, and the repeated parts will not be elaborated.

[0096] Figure 7 It is a schematic diagram of the dynamic measurement device for the reduced value of the switch nose rail in an embodiment of the present invention, as Figure 7 shown. The device 700 includes:

[0097] A data acquisition module 701, configured to acquire switch point cloud data;

[0098] Segmentation model processing module 702 is used to classify and identify the turnout point cloud data using a pre-trained multi-scale semantic segmentation model, and output point cloud data of the base rail and multiple turnout structural components; the turnout structural components include wing rails, point rails, and point rails; the multi-scale semantic segmentation model is trained based on PointNet++;

[0099] The reduction value calculation module 703 is used to fit the point cloud data of the base rail and wing rail based on the least squares method to obtain the rail top surface curve and rail head side curve; obtain the rail vertex and track gauge point based on the rail top surface curve and rail head side curve; determine the reduction value feature point based on the rail head side curve according to the geometric structure relationship between the base rail and the point rail; the reduction value feature point reflects the vertex of the point rail or the point rail; and determine the reduction value of the point rail and the fixed point rail based on the height difference between the rail vertex and the reduction value feature point.

[0100] In one embodiment, the multi-scale semantic segmentation model includes a sampling layer, a grouping layer, a PointNet layer, a feature propagation layer, and an output layer;

[0101] The sampling layer is used to downsample the input turnout point cloud data using the FPS method;

[0102] The grouping layer is used to construct local area point sets for the downsampled turnout point cloud data based on the K-nearest neighbor algorithm;

[0103] The PointNet layer is used to extract the features of each local area point set;

[0104] The feature propagation layer is used to propagate the features of the local area point set to the original point through interpolation, and output the features that integrate multi-level information;

[0105] The output layer is used to: use the features of multi-level information fusion to perform classification and recognition, and output the turnout structural components corresponding to each point.

[0106] In one embodiment, the sampling layer is specifically used to:

[0107] The input turnout point cloud data is divided into multiple sub-areas; the sub-areas include the rail head area, the rail waist area and the area below the rail bottom;

[0108] Set the downsampling rate of each sub-region according to the point cloud ratio of the sub-region and the turnout point cloud data;

[0109] According to the downsampling rate of each sub-region, the FPS method is used to downsample each sub-region.

[0110] In one embodiment, the reduction value calculation module 703 is specifically configured to:

[0111] The point cloud data of the base rail or the wing rail is used to filter and obtain the characteristic points of the rail top surface contour curve;

[0112] Using the standard equation of a circle, an objective function based on the least squares principle is constructed; the objective function includes the parameters to be solved;

[0113] The gradient descent method is used to iteratively adjust the parameters to be solved to minimize the objective function.

[0114] In one embodiment, the reduction value calculation module 703 is specifically configured to:

[0115] The extreme point of the second-order derivative of the rail head side curve is calculated; the extreme point reflects the close contact position between the stock rail and the point rail;

[0116] Calculate the normal line L of the extreme point on the side curve of the rail head I ;

[0117] Determine the distance normal L in the set S I The nearest point R; the set S is the set of base rails and point rails in the turnout point cloud data, or the set of wing rails and center rails in the turnout point cloud data;

[0118] Starting from point R, search backward for the point T with the largest vertical coordinate value in the gauge point setting area, and record it as the reduced value characteristic point.

[0119] An embodiment of the present invention also 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 computer program, the above-mentioned dynamic measurement method of the turnout point rail lowering value is implemented.

[0120] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for dynamically measuring the lowering value of the turnout point rail is implemented.

[0121] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned dynamic measurement method of the turnout point rail lowering value.

[0122] In this embodiment of the present invention, the PointNet++ model is first used to classify and identify turnout point cloud data, outputting point cloud data for multiple turnout structural components. This improves the accuracy of component identification in turnout tip rail scenarios. An adaptive least-squares fitting algorithm is then used to extract rail vertices, gauge points, and drop feature points, enabling high-precision measurement of tip rail drop and fixed point rail drop. Experimental verification demonstrates that this method achieves a measurement accuracy of ±0.3mm under dynamic detection conditions, meeting the millimeter-level detection standard for high-speed railway turnouts and improving the accuracy and efficiency of turnout tip rail drop measurement.

[0123] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0127] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic measurement method for the reduction value of the turnout point rail, characterized in that: include: Get turnout point cloud data; Using a pre-trained multi-scale semantic segmentation model trained based on PointNet++, the turnout point cloud data is classified and identified, and point cloud data of the base rail and multiple turnout structural components are output. The turnout structural components include wing rails, point rails, and point rails. Based on the least squares method, the point cloud data of the base rail and wing rail are used to fit the rail top surface curve and rail head side surface curve respectively. Based on the rail top curve and rail head side curve, the rail vertex and rail gauge point are obtained; According to the geometric structure relationship between the stock rail and the point rail, the reduction value characteristic point is determined based on the side curve of the rail head; The reduced value feature points reflect the apex of the point rail or the center rail; The height difference between the rail vertex and the reduction value feature point is used to determine the reduction value of the point rail and the fixed center rail.

2. The method according to claim 1, wherein The multi-scale semantic segmentation model includes a sampling layer, a grouping layer, a PointNet layer, a feature propagation layer and an output layer; The sampling layer is used to downsample the input turnout point cloud data using the farthest point sampling FPS method; The grouping layer is used to construct local area point sets for the downsampled turnout point cloud data based on the K-nearest neighbor algorithm; The PointNet layer is used to extract the features of each local area point set; The feature propagation layer is used to propagate the features of the local area point set to the original point through interpolation, and output the features that integrate multi-level information; The output layer is used to: use the features of multi-level information fusion to perform classification and recognition, and output the turnout structural components corresponding to each point.

3. The method according to claim 2, wherein The sampling layer is specifically used for: The input turnout point cloud data is divided into multiple sub-areas; the sub-areas include the rail head area, the rail waist area and the area below the rail bottom; Set the downsampling rate of each sub-region according to the point cloud ratio of the sub-region and the turnout point cloud data; According to the downsampling rate of each sub-region, the FPS method is used to downsample each sub-region.

4. The method according to claim 1, wherein Based on the least squares method, the point cloud data of the base rail and wing rail are fitted to obtain the rail top surface curve and rail head side curve, including: The point cloud data of the base rail or the wing rail is used to filter and obtain the characteristic points of the rail top surface contour curve; Using the standard equation of a circle, an objective function based on the least squares principle is constructed; the objective function includes the parameters to be solved; The gradient descent method is used to iteratively adjust the parameters to be solved to minimize the objective function.

5. The method according to claim 1, wherein According to the geometric structure relationship between the stock rail and the point rail, the reduction value characteristic points are determined based on the side curve of the rail head, including: The extreme point of the second-order derivative of the rail head side curve is calculated; the extreme point reflects the close contact position between the stock rail and the point rail; Calculate the normal line L of the extreme point on the side curve of the rail head I ; Determine the distance normal L in the set S I The nearest point R; the set S is the set of base rails and point rails in the turnout point cloud data, or the set of wing rails and center rails in the turnout point cloud data; Starting from point R, search backward for the point T with the largest vertical coordinate value in the gauge point setting area, and record it as the reduced value characteristic point.

6. A dynamic measuring device for the reduction value of the turnout point rail, characterized in that: include: Data acquisition module, used to obtain turnout point cloud data; A segmentation model processing module is used to classify and identify turnout point cloud data using a pre-trained multi-scale semantic segmentation model, and output point cloud data of the base rail and multiple turnout structural components; the turnout structural components include wing rails, point rails, and point rails; the multi-scale semantic segmentation model is trained based on PointNet++; The reduction value calculation module is used to obtain the rail top surface curve and rail head side curve by fitting the point cloud data of the base rail and wing rail respectively based on the least squares method; Based on the rail top curve and rail head side curve, the rail vertex and gauge point are obtained; according to the geometric structure relationship between the stock rail and the point rail, the reduction value feature point is determined based on the rail head side curve; The reduction value characteristic point reflects the apex of the point rail or the center rail; the height difference between the rail apex and the reduction value characteristic point is used to determine the reduction value of the point rail and the reduction value of the fixed center rail.

7. The device according to claim 6, characterized in that The multi-scale semantic segmentation model includes a sampling layer, a grouping layer, a PointNet layer, a feature propagation layer and an output layer; The sampling layer is used to downsample the input turnout point cloud data using the FPS method; The grouping layer is used to construct local area point sets for the downsampled turnout point cloud data based on the K-nearest neighbor algorithm; The PointNet layer is used to extract the features of each local area point set; The feature propagation layer is used to propagate the features of the local area point set to the original point through interpolation, and output the features that integrate multi-level information; The output layer is used to: use the features of multi-level information fusion to perform classification and recognition, and output the turnout structural components corresponding to each point.

8. The device according to claim 7, wherein The sampling layer is specifically used for: The input turnout point cloud data is divided into multiple sub-areas; the sub-areas include the rail head area, the rail waist area and the area below the rail bottom; Set the downsampling rate of each sub-region according to the point cloud ratio of the sub-region and the turnout point cloud data; According to the downsampling rate of each sub-region, the FPS method is used to downsample each sub-region.

9. The device according to claim 6, wherein The reduction value calculation module is specifically used for: The point cloud data of the base rail or the wing rail is used to filter and obtain the characteristic points of the rail top surface contour curve; Using the standard equation of a circle, an objective function based on the least squares principle is constructed; the objective function includes the parameters to be solved; The gradient descent method is used to iteratively adjust the parameters to be solved to minimize the objective function.

10. The device according to claim 6, wherein The reduction value calculation module is specifically used for: The extreme point of the second-order derivative of the rail head side curve is calculated; the extreme point reflects the close contact position between the stock rail and the point rail; Calculate the normal line L of the extreme point on the side curve of the rail head I ; Determine the distance normal L in the set S I The nearest point R; the set S is the set of base rails and point rails in the turnout point cloud data, or the set of wing rails and center rails in the turnout point cloud data; Starting from point R, search backward for the point T with the largest vertical coordinate value in the gauge point setting area, and record it as the reduced value characteristic point.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images