A method and system for extracting railway rail vertex cloud data based on deep learning

Abstract

The present invention discloses a method and system for extracting railway rail top point cloud data based on deep learning. The method includes: classifying the track point cloud data using a point cloud network enhanced multi-scale classification model to obtain a track key component point cloud classification data set; extracting the rail head point cloud data from the track key component point cloud classification data set, performing center line fitting to generate a rail head center line, and selecting reference points on the rail head center line; adopting a rolling window method to generate a number of slices along the rail head center line according to the reference points; performing rail top point extraction on the number of slices to obtain a continuous set of rail top coordinate points. The system includes a data classification module, a rail head center line generation module, a slice generation module, and a rail top point extraction module. The present invention can improve the efficiency of rail top data extraction, ensure the accuracy of the data and the safety of the track, and contribute to the scientific maintenance and management of railway tracks.

Description

Technical Field

[0001] The present invention relates to the technical field of railway engineering, and particularly relates to a method and system for extracting railway rail top point cloud data based on deep learning. Background Art

[0002] In recent years, with the continuous upgrading and expansion of China's railway infrastructure, especially the construction and operation of high-speed railways and heavy-haul railways, the operating quality and safety of railway lines have been subject to increasingly strict requirements. The rail top part in the track structure is a key component in direct contact with the train, and its state has an important impact on the running safety, stability and comfort of the train. However, due to the long-term service of the track and the frequent operation of the train, the track is prone to varying degrees of deformation and wear, which poses higher technical challenges to railway maintenance.

[0003] The acquisition of track point cloud data mainly relies on various measurement means such as laser scanning, satellite positioning and inertial measurement. These technologies can accurately obtain the three-dimensional coordinate information of the track. However, during the track measurement process, affected by various factors, such as unstable satellite signals, equipment vibration during laser measurement, and the complexity of the track environment, some data often have noise and errors. If these data are not effectively processed, the error points will seriously affect the accuracy of rail top extraction, and further affect subsequent track maintenance and deformation analysis.

[0004] In order to ensure the safety and operation quality of railway tracks, it is necessary to accurately extract the rail top coordinate points of the track for analyzing the deformation of the track. Traditional rail top extraction methods rely on manual inspection and simple mathematical models. These methods are inefficient and inaccurate when dealing with large-scale and highly complex track point cloud data. With the rapid development of deep learning technology, point cloud data processing methods based on deep learning have gradually been applied to track engineering, and can analyze the track more automatically and accurately. Summary of the Invention

[0005] In view of this, the purpose of the embodiments of the present invention is to provide a method and system for extracting railway rail top point cloud data based on deep learning, which can improve the efficiency of rail top data extraction, ensure the accuracy of the data and the safety of the track, and contribute to the scientific maintenance and management of railway tracks.

[0006] In a first aspect, the embodiments of the present invention provide an extraction of railway rail top point cloud data based on deep learning, which includes:

[0007] Classify the track point cloud data using a point cloud network enhanced multi-scale classification model to obtain a track key component point cloud classification data set.

[0008] Extract the rail head point cloud data from the rail key component point cloud classification dataset, perform center line fitting to generate the rail head center line, and select reference points on the rail head center line.

[0009] Adopt the rolling window method to generate a number of slices along the rail head center line according to the reference points.

[0010] Extract the rail top vertices from a number of the slices to obtain a continuous set of rail top coordinate points.

[0011] Combined with the first aspect, the embodiments of the present invention provide the first possible implementation manner of the first aspect, wherein the using the point cloud network enhanced multi-scale classification model to classify the rail point cloud data to obtain the rail key component point cloud classification dataset includes:

[0012] Collect through a laser scanning device to obtain discrete rail point cloud data, and input the discrete rail point cloud data into the point cloud network enhanced multi-scale classification model.

[0013] Perform multi-scale feature extraction on the discrete rail point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different rail key components.

[0014] Divide the discrete rail point cloud data into point cloud data of several categories, and the categories include at least one of a rail head, a rail body, fasteners, and a track slab.

[0015] Aggregate the point cloud data of different categories respectively to obtain the rail key component point cloud classification dataset.

[0016] Combined with the first aspect, the embodiments of the present invention provide the second possible implementation manner of the first aspect, wherein the performing multi-scale feature extraction on the discrete rail point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different rail key components includes:

[0017] For each point in the point cloud , define a neighborhood point set .

[0018] Calculate the geometric features of each point and its neighborhood points , and extract local features , wherein is the feature calculation result of the neighborhood point to the point .

[0019] Combined with the first aspect, the embodiments of the present invention provide a third possible implementation manner of the first aspect. Among them, extracting the rail head point cloud data from the rail key component point cloud classification dataset, performing center line fitting to generate a rail head center line, and selecting reference points on the rail head center line include:

[0020] Extract the rail head point cloud data from the rail key component point cloud classification dataset, and the rail head point cloud data includes the spatial coordinates (X, Y, Z) of the rail head.

[0021] Use the locally weighted regression method to assign different weights to the point cloud data to capture the local features of the rail head.

[0022] Use the multi - spline interpolation method to interpolate the fitting points to generate a continuous rail head center line.

[0023] On the rail head center line, select equally spaced points as reference points.

[0024] Combined with the first aspect, the embodiments of the present invention provide a fourth possible implementation manner of the first aspect. Among them, using the locally weighted regression method to assign different weights to the point cloud data to capture the local features of the rail head includes:

[0025] For each fitting point of the rail head point cloud data, use the fitting function Calculate the weights of its neighborhood points, where Is the smoothing parameter, which is used to control the weighted range of neighborhood points.

[0026] Combined with the first aspect, the embodiments of the present invention provide a fifth possible implementation manner of the first aspect. Among them, using the multi - spline interpolation method to interpolate the fitting points to generate a continuous rail head center line includes:

[0027] Let the discrete fitting point set of the rail head center line be , and each point Represents the specific spatial coordinates on the rail head center line. Take each point As an interpolation node.

[0028] In the interval between every two adjacent interpolation nodes Define the cubic spline function , where Are the coefficients of the i - th segment of the spline, and each segment of the spline Is valid in the interval .

[0029] Set the interpolation conditions, and the value of each segment of the spline at each node Is equal to the known fitting point coordinate .

[0030] Set the continuity condition of the first derivative. At the node , the slope of the spline in the previous segment is equal to the slope of the spline in the subsequent segment .

[0031] Set the continuity condition of the second derivative. The curvature of the spline in the previous segment is equal to the curvature of the spline in the subsequent segment ;

[0032] Set the natural boundary condition. The second derivative of the cubic spline function curve is zero at both ends ;

[0033] Construct the interpolation condition, the continuity condition of the first derivative, the continuity condition of the second derivative, and the natural boundary condition into a linear equation, and solve to obtain the coefficients of each segment of the spline .

[0034] Combine each segment of the spline to obtain a continuous rail head center line.

[0035] Combined with the first aspect, the embodiment of the present invention provides a sixth possible implementation manner of the first aspect, wherein, the using the rolling window method to generate a plurality of slices along the rail head center line according to the reference point includes:

[0036] Generate a plurality of main slices on the rail head center line according to the selected reference point.

[0037] Through the formula generate subdivision slices between adjacent main slices, wherein, is the normal vector of the subdivision slice plane, parallel or perpendicular to the direction vector of the rail head center line, is the position vector of a point on the rail head center line, the center of the subdivision slice at this position, is the position vector of any point within the subdivision slice.

[0038] Rollingly generate the subdivision slices between the main slices along the rail head center line in sequence.

[0039] Combined with the first aspect, the embodiment of the present invention provides a seventh possible implementation manner of the first aspect, wherein, the extracting the rail vertex points from a plurality of the slices to obtain a continuous set of rail top coordinate points includes:

[0040] For the point cloud data in each slice, select the point with the highest height value as the rail vertex of the slice according to the height value of each point.

[0041] Arrange the coordinates of the rail vertices of each slice in the order of the slices on the center line of the rail head in the railway direction to form the rail top line.

[0042] Connect the coordinates of all the extracted rail vertices in sequence to generate a continuous set of rail top coordinate points.

[0043] Combined with the first aspect, the embodiment of the present invention provides an eighth possible implementation manner of the first aspect, wherein, for the point cloud data in each slice, according to the height value of each point, selecting the point with the highest height as the rail vertex of the slice includes:

[0044] For each slice, traverse all the point cloud data to obtain the height value of each point .

[0045] Define the height of the point corresponding to the highest height value as the rail top height ,

[0046] Record the three-dimensional coordinates (X, Y, Z) of this point as the rail vertex of the slice.

[0047] In the second aspect, the embodiment of the present invention also provides a system for extracting railway rail vertex point cloud data based on deep learning, which includes:

[0048] A data classification module for classifying the track point cloud data using a point cloud network enhanced multi-scale classification model to obtain a track key component point cloud classification data set.

[0049] A rail head center line generation module for extracting the rail head point cloud data from the track key component point cloud classification data set, performing center line fitting, generating a rail head center line, and selecting reference points on the rail head center line.

[0050] A slice generation module for using the rolling window method to generate a number of slices along the rail head center line according to the reference points.

[0051] A rail vertex extraction module for extracting rail vertices from a number of the slices to obtain a continuous set of rail top coordinate points.

[0052] The beneficial effects of the present invention are:

[0053] The present invention provides a fully automated method for extracting rail head vertices, which does not rely on manual intervention, greatly reducing the necessity of manual participation and the possibility of human error. Through the automatic loading of point cloud data, center line fitting, and vertex extraction, the entire process is highly integrated and automated, making the track detection process more efficient and fast, meeting the requirements of modern railway maintenance for high efficiency.

[0054] Using local weighted regression and multiple spline interpolation to fit the center line of the rail head point cloud data helps to ensure the accuracy of the center line and reduces the deviation caused by point cloud noise and abnormal data. At the same time, the fixed-interval slice analysis method can accurately identify the highest point in each slice as the rail vertex, ensuring the height accuracy and reliability of the extracted points. This high-precision extraction method effectively improves the accuracy of track flatness evaluation and deformation monitoring.

[0055] The present invention ensures that the extraction of rail vertices has the characteristic of uniform distribution in the track direction by the method of slicing at fixed intervals along the center line. The uniform vertex extraction is crucial for the flatness evaluation of the track and can effectively avoid the deviation of track condition evaluation caused by uneven sampling. The uniformly distributed vertex data provides a scientific basis for the long-term monitoring and maintenance of the track, can more comprehensively reflect the actual state of the track, and supports the accurate evaluation of track deformation.

[0056] The method of the present invention has strong applicability and is not only applicable to the measurement of different track shapes and lengths, but also can cope with the accuracy and density of various point cloud data. By flexibly adjusting the number of center line fitting points and the interval distance of slice analysis, it can adapt to different types and conditions of track detection tasks and has broad application prospects.

[0057] The present invention provides a scientific and effective method for the automatic extraction of rail head vertices. The research results are of great significance for the detection and maintenance of railway tracks. Through the highly automated and high-precision extraction method, the efficiency and quality of track measurement and evaluation are greatly improved, providing important technical support for the healthy and safe development of the railway industry. Description of the Drawings

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0059] Figure 1 It is the flowchart of the method for extracting railway rail head point cloud data based on deep learning of the present invention;

[0060] Figure 2 It is the classification schematic diagram of the track point cloud data in the method for extracting railway rail head point cloud data based on deep learning of the present invention;

[0061] Figure 3 It is the schematic diagram of multiple spline interpolation of the rail head center line in the method for extracting railway rail head point cloud data based on deep learning of the present invention;

[0062] Figure 4 This is a schematic diagram of generating slices along the center line of the rail head in the method for extracting railway rail vertex cloud data based on deep learning of the present invention;

[0063] Figure 5 This is a schematic diagram of extracting rail vertexes within the cross-sectional slice in the method for extracting railway rail vertex cloud data based on deep learning of the present invention;

[0064] Figure 6 This is a schematic diagram of generating the rail top coordinate point set and the continuous rail top linear shape in the method for extracting railway rail vertex cloud data based on deep learning of the present invention;

[0065] Figure 7 This is a schematic diagram of the extraction process of the method for extracting railway rail vertex cloud data based on deep learning of the present invention. Specific implementation manners

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0067] The present invention proposes a method for extracting railway rail vertex cloud data based on deep learning and rolling window slicing. The PointNet++ MSG model is used to classify the track point cloud, accurately extract the data of the rail head part, and combine the center line fitting and rolling window slicing analysis technologies to ensure the high-precision extraction of rail vertexes. This method not only improves the efficiency of rail top data extraction but also ensures the accuracy of the data and the safety of the track, which is helpful for the scientific maintenance and management of railway tracks.

[0068] Please refer to Figures 1 to 7 , the first embodiment of the present invention provides a method for extracting railway rail vertex based on deep learning, which includes: using a point cloud network enhanced multi-scale classification model to classify the track point cloud data to obtain a track key component point cloud classification data set; extracting the rail head point cloud data from the track key component point cloud classification data set, performing center line fitting to generate a rail head center line, and selecting reference points on the rail head center line; adopting the rolling window method to generate a plurality of slices along the rail head center line according to the reference points; and extracting rail vertexes from the plurality of slices to obtain a continuous rail top coordinate point set.

[0069] Among them, using the point cloud network enhanced multi-scale classification model to classify the track point cloud data to obtain the track key component point cloud classification data set includes: collecting through a laser scanning device to obtain discrete track point cloud data, and inputting the discrete track point cloud data into the point cloud network enhanced multi-scale classification model; performing multi-scale feature extraction on the discrete track point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different track key components; dividing the discrete track point cloud data into point cloud data of several categories, where the categories include at least one of a rail head, a rail body, fasteners, and a track slab; aggregating the point cloud data of different categories respectively to obtain the track key component point cloud classification data set.

[0070] Specifically, the rail head has a flat top geometry, and the point cloud network enhanced multi-scale classification model distinguishes it by identifying the horizontal structure features through the neighborhood point relationship; the rail body is long and vertical, adjacent to the rail head, and is distinguished by its longitudinal features; the track slab is distinguished by extracting its planarity through multi-scale information; the fasteners are smaller in volume than the rail head, the rail body, and the track slab, and are distinguished by being close to the position of the track slab but having a relatively independent shape.

[0071] Among them, performing multi-scale feature extraction on the discrete track point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different track key components includes: for each point in the point cloud , defining a neighborhood point set ; calculating the geometric features of each point and its neighborhood points to extract local features , where is the feature calculation result of the neighborhood point to the point . By aggregating these local features, it is possible to better classify each component of the track, as shown in Figure 2 .

[0072] Among them, extracting the rail head point cloud data from the track key component point cloud classification data set, performing center line fitting to generate a rail head center line, and selecting reference points on the rail head center line includes: extracting the rail head point cloud data from the track key component point cloud classification data set, where the rail head point cloud data contains the spatial coordinates (X, Y, Z) of the rail head; using the local weighted regression method to assign different weights to the point cloud data to capture the local features of the rail head; using the multi-spline interpolation method to interpolate the fitting points to generate a continuous rail head center line; on the rail head center line, selecting equally spaced points as reference points.

[0073] Among them, using the locally weighted regression method to assign different weights to the point cloud data and capture the local features of the rail head includes: for each fitting point of the rail head point cloud data, using a fitting function to calculate the weights of its neighborhood points, where is a smoothing parameter used to control the weighted range of neighborhood points, ensuring that points closer to the center point have a greater impact on the fitting result.

[0074] Among them, using the multiple spline interpolation method to interpolate the fitting points to generate a continuous rail head center line includes: setting the discrete fitting point set of the rail head center line as , and each point represents the specific spatial coordinates on the rail head center line. Taking each point as an interpolation node to form the demarcation points of multiple segments of splines; defining a cubic spline function in the interval between every two adjacent interpolation nodes, where are the coefficients of the i-th segment of the spline, and each segment of the spline is valid in the interval ; setting interpolation conditions, the value of each segment of the spline at each node is equal to the known fitting point coordinate , ensuring that the function values of each segment of the spline are consistent with the discrete fitting points of the rail head at the interpolation nodes; setting the continuity condition of the first derivative, at the node , the slope of the previous segment of the spline is equal to the slope of the next segment of the spline ; setting the continuity condition of the second derivative, the curvature of the previous segment of the spline is equal to the curvature of the next segment of the spline ; setting natural boundary conditions, the second derivative of the cubic spline function curve is zero at both ends , this condition makes the change rate of the curve smooth at the boundaries, and the obtained interpolation curve smoothly transitions at both ends of the rail head center line; constructing the interpolation conditions, the continuity condition of the first derivative, the continuity condition of the second derivative, and the natural boundary conditions into a linear equation, and solving to obtain the coefficients of each segment of the spline; combining each segment of the spline to obtain a continuous rail head center line, as shown in Figure 3 . In the interval , the rail head center line is spliced by these segmented splines, and it satisfies the requirements of smoothness and continuity in each interpolation node and its interval area.

[0075] Among them, the rolling window method is adopted. Generating a number of slices along the center line of the rail head according to the reference points includes: generating a number of main slices on the center line of the rail head according to the selected reference points. The reference points are evenly distributed, and the interval distance can be adjusted according to actual needs to ensure the smooth transition of the track and the data density; through the formula Generating subdivision slices between adjacent main slices, where is the normal vector of the subdivision slice plane, parallel or perpendicular to the direction vector of the center line of the rail head, is the position vector of a certain point on the center line of the rail head, which is the center of the subdivision slice at this position, is the position vector of any point within the subdivision slice. The generated subdivision slices will be parallel to the local plane of the rail head at this point, ensuring that the subdivision slices can be perpendicular to the center line of the rail head, thereby maximizing the consistency of the track direction. The number and spacing of the subdivision slices can be set according to the requirements of rail top feature extraction. Usually, dense subdivision slices are selected to increase the resolution of the rail top data. As Figure 4 shown, by generating a number of slices between every two center line points, the extraction density of the rail top coordinates can be effectively increased, thereby ensuring that the rail top data has high resolution and high continuity; rolling and sequentially generating the subdivision slices between the main slices along the center line of the rail head. The moving step size and window size of the rolling window can be adjusted according to the track shape and measurement accuracy requirements. Each generated subdivision slice not only contains local rail head geometric information but also forms a continuous rail top data set with adjacent slices, providing a complete track top surface contour.

[0076] Each generated slice serves as an independent planar region for further analyzing the geometric features of the rail head point cloud. The way of slice generation ensures a detailed analysis of the rail head at different positions, thus providing a deeper understanding of the details along the track. This process provides a data basis for the accurate extraction of the rail top.

[0077] Among them, extracting rail top points from a number of the slices to obtain a continuous set of rail top coordinate points includes: for the point cloud data in each slice, according to the height value of each point, selecting the point with the highest height as the rail top point of the slice; arranging the coordinates of the rail top points of each slice in the order of the slices on the center line of the rail head in the railway direction to form a rail top line; connecting the coordinates of all the extracted rail top points in sequence to generate a continuous set of rail top coordinate points. The sequential arrangement of the rail top coordinate point set ensures the continuity and accuracy of the rail top data, thereby accurately reflecting the shape of the track. As Figure 6 shown.

[0078] Among them, for the point cloud data in each of the said slices, selecting the point with the highest height as the rail vertex of the said slice according to the height value of each point includes: for each of the said slices, traversing all the point cloud data to obtain the height value of each point ; defining the height of the point corresponding to the highest of the said height values as the rail top height , recording the three-dimensional coordinates (X, Y, Z) of this point as the rail vertex of the said slice. As Figure 5 shown, by comparing the heights of all points in each slice, the point with the maximum height is extracted, thereby ensuring the accuracy of the rail vertex.

[0079] The second embodiment of the present invention provides a railway rail vertex cloud data extraction system based on deep learning, which includes: a data classification module for classifying track point cloud data using a point cloud network enhanced multi-scale classification model to obtain a track key component point cloud classification data set; a rail head center line generation module for extracting rail head point cloud data from the track key component point cloud classification data set, performing center line fitting to generate a rail head center line, and selecting reference points on the rail head center line; a slice generation module for using a rolling window method to generate a number of slices along the rail head center line according to the reference points; a rail vertex extraction module for extracting rail vertices from the number of said slices to obtain a continuous set of rail top coordinate points.

[0080] The embodiments of the present invention aim to protect a method and system for extracting railway rail vertex cloud data based on deep learning, and have the following effects:

[0081] In response to the current demand for extracting rail top vertices of railway tracks, the present invention proposes an automated rail top vertex extraction method based on deep learning and rolling window slicing. First, rail head point cloud data is obtained through deep learning, and then the center line of the rail head point cloud data is fitted through local weighted regression and multiple spline interpolations to ensure the accuracy of the center line; finally, rail vertices are extracted based on slice analysis at fixed intervals to ensure that the rail vertices are evenly distributed along the track direction.

[0082] The computer program product of the method and device for extracting railway rail vertex cloud data based on deep learning provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods in the foregoing method embodiments. For specific implementation, reference can be made to the method embodiments, which will not be elaborated here.

[0083] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned method for extracting railway rail vertex cloud data based on deep learning, thereby improving the efficiency of rail top data extraction and ensuring the accuracy of the data and the safety of the track.

[0084] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.

[0085] Finally, it should be noted that: the above-mentioned embodiments are only specific implementation manners of the present invention, used to illustrate the technical solution of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A railway track vertex cloud data extraction method based on deep learning, characterized in that: include: The track point cloud data is classified using the point cloud network enhanced multi-scale classification model to obtain a point cloud classification dataset of key track components. Extracting rail head point cloud data from the rail key component point cloud classification data set, performing centerline fitting, generating a rail head centerline, and selecting a reference point on the rail head centerline; Using a rolling window method, a plurality of slices are generated along the center line of the rail head according to the reference point; Extracting rail vertices from a number of the slices to obtain a continuous rail top coordinate point set; The point cloud network enhanced multi-scale classification model is used to classify the track point cloud data to obtain a point cloud classification data set of key track components, including: Acquire discrete track point cloud data through laser scanning equipment, and input the discrete track point cloud data into the point cloud network enhanced multi-scale classification model; The point cloud network enhanced multi-scale classification model is PointNet++MSG; Performing multi-scale feature extraction on the discrete track point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different key track components; Classifying the discrete track point cloud data into several categories of point cloud data, wherein the categories include at least one of a rail head, a rail body, a fastener, and a rail plate; The point cloud data of different categories are aggregated respectively to obtain a point cloud classification data set of key track components.

2. The railway track vertex cloud data extraction method based on deep learning according to claim 1 is characterized in that: The multi-scale feature extraction of the discrete track point cloud data by the point cloud network enhanced multi-scale classification model to capture the features of different key track components includes: For each point in the point cloud , define the neighborhood point set ; Calculate each of the points and its neighboring points Geometric features, extract local features ,in, is the neighborhood point To the point The characteristic calculation results of .

3. The railway track vertex cloud data extraction method based on deep learning according to claim 1 is characterized in that: The extracting of the rail head point cloud data from the rail key component point cloud classification data set, performing centerline fitting, generating the rail head centerline, and selecting a reference point on the rail head centerline comprises: Extracting rail head point cloud data from the rail key component point cloud classification data set, wherein the rail head point cloud data includes the spatial coordinates (X, Y, Z) of the rail head; The local weighted regression method is used to assign different weights to the point cloud data to capture the local features of the rail head; The fitting points are interpolated using multiple spline interpolation method to generate a continuous rail head centerline; On the center line of the rail head, equally spaced points are selected as reference points.

4. The railway track vertex cloud data extraction method based on deep learning according to claim 3 is characterized in that: The local weighted regression method is used to assign different weights to the point cloud data, and the local features of the rail head are captured, including: For each of the fitting points of the rail head point cloud data, a fitting function is used. Calculate the weight of its neighborhood points, where is a smoothing parameter used to control the weighted range of neighborhood points.

5. The railway track vertex cloud data extraction method based on deep learning according to claim 3 is characterized in that: The method of interpolating the fitting points using multiple spline interpolation method to generate a continuous rail head centerline includes: Assume that the discrete fitting point set of the rail head centerline is , each point Represents the spatial coordinates on the center line of the rail head, and each point As an interpolation node; In the interval between every two adjacent interpolation nodes Define the cubic spline function ,in, is the coefficient of the i-th spline, each spline In the said interval Internally valid; Set the interpolation conditions, each segment of the spline at each node The value at is equal to the known coordinates of the fitting point ; Set the continuity condition for the first-order derivative at the node At this point, the slope of the spline in the first section is equal to the slope of the spline in the second section. ; Set the continuity condition of the second-order derivative, the curvature of the spline described in the first paragraph is equal to the curvature of the spline described in the second paragraph ; Set natural boundary conditions, the second-order derivative of the cubic spline function curve at both ends is zero ; The interpolation condition, the continuity condition of the first-order derivative, the continuity condition of the second-order derivative and the natural boundary condition are constructed into a linear equation, and the coefficients of each segment of the spline are obtained by solving the equation. ; Each segment of the spline Combined together, a continuous rail head center line is obtained.

6. The railway track vertex cloud data extraction method based on deep learning according to claim 1, characterized in that: The method of using the rolling window method to generate a plurality of slices along the center line of the rail head according to the reference point comprises: According to the selected reference points, a plurality of main slices are generated on the center line of the rail head; By formula Subdivided slices are generated between adjacent main slices, wherein: is the normal vector of the subdivided slice plane, parallel or perpendicular to the direction vector of the centerline of the rail head, is the position vector of a point on the centerline of the rail head, and the subdivision slice is centered at that position, is the position vector of any point in the subdivided slice; The subdivided slices are generated sequentially between the main slices in a rolling manner along the center line of the rail head.

7. The railway track vertex cloud data extraction method based on deep learning according to claim 1 is characterized in that: The step of extracting rail vertices from a plurality of the slices to obtain a continuous rail top coordinate point set comprises: For each of the point cloud data in the slice, according to the height value of each point, select the point with the highest height as the vertex of the track of the slice; According to the order of the slices on the center line of the rail head, the coordinates of the rail vertex of each slice are arranged in the direction of the railway to form a rail top line; The coordinates of all the extracted rail vertices are connected in sequence to generate a continuous rail top coordinate point set.

8. The railway track vertex cloud data extraction method based on deep learning according to claim 7 is characterized in that: The step of selecting the point with the highest height as the vertex of the track of each slice according to the height value of each point in the point cloud data in each slice comprises: For each of the slices, traverse all point cloud data and obtain the height value of each point ; The height value The height of the highest corresponding point is defined as the rail top height , The three-dimensional coordinates (X, Y, Z) of the point are recorded as the vertex of the track of the slice.

9. A railway track vertex cloud data extraction system based on deep learning, characterized in that: include: A data classification module is used to classify the track point cloud data using a point cloud network enhanced multi-scale classification model to obtain a point cloud classification dataset of key track components; A rail head centerline generation module is used to extract rail head point cloud data from the rail key component point cloud classification data set, perform centerline fitting, generate the rail head centerline, and select a reference point on the rail head centerline; A slice generation module, used to generate a plurality of slices along the center line of the rail head according to the reference point by using a rolling window method; A rail vertex extraction module, used to extract rail vertices from a plurality of the slices to obtain a continuous rail top coordinate point set; The operations performed by the data classification module include: Acquire discrete track point cloud data through laser scanning equipment, and input the discrete track point cloud data into the point cloud network enhanced multi-scale classification model; The point cloud network enhanced multi-scale classification model is PointNet++MSG; Performing multi-scale feature extraction on the discrete track point cloud data through the point cloud network enhanced multi-scale classification model to capture the features of different key track components; Classifying the discrete track point cloud data into several categories of point cloud data, wherein the categories include at least one of a rail head, a rail body, a fastener, and a rail plate; The point cloud data of different categories are aggregated respectively to obtain a point cloud classification data set of key track components.

Images