Railway centerline automatic extraction method and system based on airborne point cloud

Abstract

The present application relates to a kind of railway center line full-automatic extraction method and system based on airborne point cloud.The problem that there is point cloud sparseness in airborne laser point cloud re-measurement railway line needs to rely on manual marking manual extraction railway center line is solved.The method obtains point cloud data, obtains two-dimensional intensity image after plane projection, marks railway track line, and is used as training data in pairs;Convolutional neural network model RWNet is constructed, and training data is input to train;Two-dimensional intensity image is input into the model after training, and binary image of railway center line is predicted and extracted two-dimensional railway center line;Read point cloud data, search current railway current position elevation from point cloud data in the preset range along the vertical direction of railway center line;Railway three-dimensional line is updated, and railway center line is extracted based on railway three-dimensional line.The method can realize full-automatic extraction of railway center line in the case of relatively sparse point cloud data, improve efficiency, and also guarantee the accuracy of result.

Description

Technical Field

[0001] This invention relates to the field of surveying and remote sensing technology, specifically to a fully automatic method and system for extracting railway centerlines based on airborne point clouds. Background Technology

[0002] With the rapid development of railway construction, a large number of existing railways need to be upgraded to improve their operating speed and capacity, and measuring the centerline of existing railways is one of the important preliminary tasks for railway upgrading.

[0003] Railway centerline measurements are mostly conducted manually, using total stations and track trolleys with prisms mounted for point-by-point measurements. This method is labor-intensive, inefficient, and costly, and can only be performed during maintenance windows, making it unsuitable for current railway construction requirements. To overcome these issues, non-contact 3D laser scanning technology is gradually being applied in railway surveying and design. Existing schemes for railway centerline measurements using 3D laser scanning typically involve fixed-station or mobile 3D laser measurements to acquire high-density, high-precision point cloud data, which is then used to automatically detect and reconstruct the railway track. However, this method still requires maintenance windows, limiting operational time. In recent years, with the development of UAV technology, airborne laser point cloud technology has begun to be applied in railway line resurvey projects. This method overcomes the limitations of maintenance windows and is more efficient; however, airborne point cloud data is sparser than station-based 3D laser scanning, requiring manual annotation and extraction of the railway centerline.

[0004] Therefore, it is necessary to propose new methods to overcome the shortcomings of current methods for railway line resurvey using airborne laser point cloud technology. Summary of the Invention

[0005] The purpose of this invention is to provide a fully automatic method and system for extracting railway centerlines based on airborne point clouds, so as to solve the problems of sparse point cloud data and the need for manual annotation and extraction of railway centerlines in the current railway line resurvey using airborne laser point cloud technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A fully automated method for extracting railway centerlines based on airborne point clouds, the method comprising:

[0008] Obtain point cloud data and then project it onto a plane to obtain a two-dimensional intensity image;

[0009] Railway track lines are labeled in two-dimensional intensity images, and the two-dimensional intensity images and railway track lines are used as training data in pairs.

[0010] A convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images is constructed. The model is trained by inputting training data to obtain the trained convolutional neural network model RWNet.

[0011] The two-dimensional intensity image is input into the trained convolutional neural network model RWNet to predict the binary image of the rail centerline and extract the two-dimensional rail centerline, which is then stored as a node.

[0012] Read the point cloud data. For each node on the centerline of the railway track, search for the current elevation of the current position of the railway track in the point cloud data along a preset range perpendicular to the centerline of the railway track.

[0013] Update the node elevation to obtain the 3D lines of the railway track, and extract the railway centerline based on the 3D lines of the railway track.

[0014] Further, point cloud data is acquired, and a two-dimensional intensity image is obtained after planar projection, including:

[0015] Read point cloud data, the range of which is (x min x max y min y max );

[0016] The point cloud data is projected onto the xoy plane to form a two-dimensional intensity image I after planar projection;

[0017] The two-dimensional intensity image I after planar projection is divided into a grid, where each grid p represents a pixel in the two-dimensional intensity image I. The width of grid p is s, and the width w and height h of the two-dimensional intensity image I are:

[0018] h=(y max -y min ) / s

[0019] w = (w max -w min ) / s

[0020] The pixel value I of each grid p p The average intensity of the point cloud falling within this grid:

[0021]

[0022] in:

[0023] n is the number of point clouds projected into the grid p;

[0024] Intensity i The intensity value for each point cloud.

[0025] Furthermore, the two-dimensional intensity image I is normalized.

[0026] Furthermore, a convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images is constructed, including:

[0027] The convolutional neural network model RWNet uses an encoder-decoder structure;

[0028] The feature extraction structure of ResNet residual network is used as the encoding structure, and the decoder of semantic segmentation model is used as the decoding structure.

[0029] Furthermore, the decoding structure includes four upsampling layers with a sampling interval of 2, used to restore the features generated by the encoding structure to the same size as the original input intensity image;

[0030] After each upsampling layer, two convolutional layers are used to smooth the upsampling results, and the final output is a binary image of the railway centerline with the same size as the input image.

[0031] Furthermore, the two-dimensional intensity image is input into the trained convolutional neural network model RWNet to predict the binary image of the rail centerline and extract the two-dimensional rail centerline, including:

[0032] The RWNet convolutional neural network model predicts binary images of the railway track centerline;

[0033] The two-dimensional rail centerline is extracted by vectorizing the binary image of the rail centerline.

[0034] Furthermore, the railway centerline is stored in the form of nodes, including:

[0035] After vectorizing the left and right rail lines, interpolate nodes at the same spacing.

[0036] Save the lines as nodes, with a node spacing of S.

[0037] Furthermore, the current elevation of the railway track is a local elevation maximum.

[0038] Furthermore, the node elevations are updated to obtain the 3D lines of the railway track. Based on these 3D lines, the railway centerline is extracted, including:

[0039] Obtain the 3D lines of the left and right railway tracks;

[0040] For the i-th node L on the three-dimensional line of the left rail i Calculate R for each node on the 3D line of the right rail. j Its straight-line distance D ij ;

[0041] Find with L i Straight-line distance D ij The smallest right rail 3D line node Rj ;

[0042] According to L i With R j The railway centerline node M was calculated. ij coordinate:

[0043] M ij =(L i +R j ) / 2;

[0044] For each node on the three-dimensional line of the left rail, repeat the calculation and connect M sequentially. ij This generates the railway centerline.

[0045] On the other hand, a fully automated railway centerline extraction system based on airborne point clouds is provided, the system being used to implement the method, including:

[0046] The planar projection module is used to acquire point cloud data and obtain a two-dimensional intensity image after planar projection.

[0047] The annotation module is used to annotate railway track lines in two-dimensional intensity images, using the two-dimensional intensity images and railway track lines as training data in pairs.

[0048] The model training module is used to build a convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images. It is trained by inputting training data to obtain the trained convolutional neural network model RWNet.

[0049] The prediction module is used to input the two-dimensional intensity image into the trained convolutional neural network model RWNet, predict the binary image of the rail centerline, extract the two-dimensional rail centerline, and save the rail centerline in the form of nodes.

[0050] The elevation extraction module is used to read point cloud data. For each node on the centerline of the railway, it searches for the current elevation of the current position of the railway in the point cloud data along a preset range perpendicular to the centerline of the railway.

[0051] The railway centerline extraction module is used to update node elevations, obtain 3D lines of the railway track, and extract the railway centerline based on these 3D lines.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] This invention provides a fully automated method and system for extracting railway centerlines based on airborne point clouds. It projects three-dimensional laser point clouds onto the xoy plane, converting them into two-dimensional intensity images. Deep learning methods are used to identify the railway track plane position within the two-dimensional intensity image. Then, the elevation information of the railway track lines is extracted by combining the point cloud data with the railway track line plane position, resulting in three-dimensional railway track lines. Finally, the position of the railway centerline is calculated based on the left and right rail lines. This method can achieve fully automated extraction of the railway centerline even when the point cloud data is relatively sparse, improving efficiency while ensuring the accuracy of the results. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0055] Figure 1 An example of a two-dimensional intensity image of a point cloud produced for this invention.

[0056] Figure 2 This is an example of a railway track line annotation produced according to the present invention. Among them, (1) is the original two-dimensional intensity image, and (2) is the railway track line annotation result.

[0057] Figure 3 This is a schematic diagram illustrating how the present invention obtains railway track elevation information.

[0058] Figure 4 This invention presents the results of extracting the railway centerline from airborne laser point cloud data. Among them, (1) and (2) are the extraction results under different scenarios.

[0059] Figure 5 This is a flowchart of the method of the present invention. Detailed Implementation

[0060] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0061] It should be noted that similar reference numerals and letters indicate similar items; therefore, once an item is defined in one embodiment, it does not need to be further defined and explained in subsequent embodiments. Furthermore, the terms "comprising" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0062] It should also be noted that although the order of steps is mentioned in the method description, in some cases, steps may be performed in a different order than that described here, and this should not be interpreted as a restriction on the order of steps.

[0063] This invention provides a fully automatic method for extracting the railway centerline based on airborne point clouds. The method first projects a three-dimensional laser point cloud onto the xoy plane and converts it into a two-dimensional intensity image. Then, it uses deep learning to identify the plane position of the railway track in the two-dimensional intensity image. Next, it combines the point cloud data with the plane position of the railway track line to extract the elevation information of the track line, thus obtaining the three-dimensional line of the railway track. Finally, it calculates the position of the railway centerline based on the left and right rail lines, thereby achieving fully automatic extraction of the railway centerline.

[0064] like Figure 5 The method includes:

[0065] S1: Acquire point cloud data, project it onto a plane to obtain a two-dimensional intensity image, including:

[0066] S101: Read point cloud data, the range of which is (x... min ,x max ,y min ,y max );

[0067] S102: Project the point cloud data onto the xoy plane to form a two-dimensional intensity image I after planar projection;

[0068] S103: Divide the two-dimensional intensity image I after planar projection into a grid, where each grid p represents a pixel in the two-dimensional intensity image I, and the width of grid p is s. The width w and height h of the two-dimensional intensity image I are:

[0069] h=(y max -y min ) / s

[0070] w = (w max -w min ) / s

[0071] S104: Pixel value I for each grid p pThe average intensity of the point cloud falling within this grid:

[0072]

[0073] in:

[0074] n is the number of point clouds projected into the grid p;

[0075] Intensity i For each point cloud intensity value, if there is no point cloud projection within the grid, then the corresponding image pixel value for that grid is 0.

[0076] S105: Normalize the two-dimensional intensity image I.

[0077] The pixel values ​​in this step are point cloud intensity values, which need to be normalized to 0-255 in the final step. The point cloud intensity image is shown below. Figure 1 As shown.

[0078] S2: This method uses deep learning to automatically extract the planar position of railway tracks. The planar positions of the tracks need to be labeled for training the deep network model. Specifically, the open-source image annotation tool Labelme is used to annotate railway track lines in 2D intensity images. The 2D intensity images and railway track lines are paired as training data. The annotation results are as follows: Figure 2 As shown.

[0079] S3: Construct a convolutional neural network model RWNet for extracting railway track lines from 2D intensity images. Train the model using training data to obtain the trained RWNet model. After training, the ability of the RWNet model to predict the planar position of the railway tracks will be significantly improved.

[0080] This method designs a convolutional neural network model, RWNet, suitable for automatically extracting railway track plane lines from point cloud intensity images. It employs an encoder-decoder structure, similar to image edge detection, taking the point cloud intensity image as input and outputting a binary image representing the extracted railway track centerline (pixels containing the track centerline have a value of 1, while background pixels have a value of 0). The feature extraction structure of a residual network, ResNet, is used as the encoding structure, and a semantic segmentation model decoder is used as the decoding structure. The decoding structure contains four upsampling layers with a sampling interval of 2, used to restore the features generated by the encoding structure to the same size as the original input intensity image. After each upsampling layer, two convolutional layers are used to smooth the upsampling results, finally outputting a binary image of the railway track centerline with the same size as the input image.

[0081] S4: Input the 2D intensity image into the trained convolutional neural network model RWNet to predict the binary image of the rail centerline and extract the 2D rail centerline. Store the rail centerline as nodes, including:

[0082] S401: The RWNet convolutional neural network model predicts a binary image of the railway track centerline;

[0083] S402: Extract the rail centerline by vectorizing the binary image of the rail centerline;

[0084] S403: Vectorize the left and right rail lines and then interpolate nodes at the same interval;

[0085] S404: Saves lines as nodes, with a node spacing of S.

[0086] S5: Read the point cloud data. Based on prior knowledge of the rail top shape (the rail top has a maximum elevation within a certain range perpendicular to the railway), for each node on the rail centerline, search for the current elevation of the current rail position in the point cloud data within a preset range perpendicular to the rail centerline. The search method is as follows: Figure 3 As shown.

[0087] The current elevation of the railway track is a local elevation maximum.

[0088] S6: Update the node elevation for each rail midpoint in this manner to obtain the 3D rail line. Extract the railway centerline based on the 3D rail line, including:

[0089] S601: Obtain the 3D lines of the left and right railway tracks;

[0090] S602: For the i-th node L on the three-dimensional line of the left rail... i Calculate R for each node on the 3D line of the right rail. j Its straight-line distance D ij ;

[0091] S603: Find L i Straight-line distance D ij The smallest right rail 3D line node R j ;

[0092] S604: According to L i With R j The railway centerline node M was calculated. ij coordinate:

[0093] M ij =(L i +R j ) / 2;

[0094] S605: For each node on the three-dimensional line of the left rail, repeat the calculation and connect M sequentially. ij This generates the railway centerline.

[0095] On the other hand, the present invention provides a fully automated railway centerline extraction system based on airborne point clouds, the system being used to implement the above-described method, including:

[0096] The planar projection module is used to acquire point cloud data and obtain a two-dimensional intensity image after planar projection, which corresponds to S1 of the above method;

[0097] The annotation module is used to annotate railway track lines in the two-dimensional intensity image. The two-dimensional intensity image and the railway track lines are paired as training data, corresponding to S2 of the above method.

[0098] The model training module is used to construct a convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images. It is trained by inputting training data to obtain the trained convolutional neural network model RWNet, which corresponds to S3 of the above method.

[0099] The prediction module is used to input the two-dimensional intensity image into the trained convolutional neural network model RWNet, predict the binary image of the rail centerline and extract the two-dimensional rail centerline, and save the rail centerline in the form of nodes, which corresponds to S4 of the above method.

[0100] The elevation extraction module is used to read point cloud data. For each node on the centerline of the railway, it searches for the current elevation of the current position of the railway in the point cloud data along a preset range perpendicular to the centerline of the railway, corresponding to S5 of the above method.

[0101] The railway centerline extraction module is used to update node elevations, obtain the three-dimensional lines of the railway track, and extract the railway centerline based on the three-dimensional lines of the railway track, corresponding to S6 of the above method.

[0102] Those skilled in the art will understand that all or part of the functions of the embodiments of the present invention can be implemented by hardware or by computer program. When all or part of the functions in the above embodiments are implemented by computer program, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the above functions are implemented by executing the program by a computer. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be implemented. In addition, when all or part of the functions in the above embodiments are implemented by computer program, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash drive, or portable hard drive, and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be implemented.

[0103] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the ideas of this invention.

Claims

1. A fully automatic method for extracting railway centerlines based on airborne point clouds, characterized in that: The method includes: Obtain point cloud data and then project it onto a plane to obtain a two-dimensional intensity image; Railway track lines are labeled in two-dimensional intensity images, and the two-dimensional intensity images and railway track lines are used as training data in pairs. A convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images is constructed. The model is trained by inputting training data to obtain the trained convolutional neural network model RWNet. The two-dimensional intensity image is input into the trained convolutional neural network model RWNet to predict the binary image of the rail centerline and extract the two-dimensional rail centerline, which is then stored as a node. Read the point cloud data. For each node on the centerline of the railway track, search for the current elevation of the current position of the railway track in the point cloud data along a preset range perpendicular to the centerline of the railway track. Update node elevations to obtain 3D lines of the railway track, and extract the railway centerline based on the 3D lines of the railway track; in: Point cloud data is acquired, and after planar projection, a two-dimensional intensity image is obtained, including: Read point cloud data; the range of the point cloud data is ( , , , ); The point cloud data is projected onto the xoy plane to form a two-dimensional intensity image I after planar projection; The two-dimensional intensity image I after planar projection is divided into a grid, where each grid p represents a pixel in the two-dimensional intensity image I. The width of grid p is s, and the width w and height h of the two-dimensional intensity image I are: Pixel values ​​of each grid p The average intensity of the point cloud falling within this grid: in: n is the number of point clouds projected into the grid p; The intensity value for each point cloud.

2. The fully automatic railway centerline extraction method based on airborne point clouds according to claim 1, characterized in that... Normalize the two-dimensional intensity image I.

3. The fully automatic railway centerline extraction method based on airborne point clouds according to claim 2, characterized in that: Constructing a convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images includes: The convolutional neural network model RWNet uses an encoder-decoder structure; The feature extraction structure of ResNet residual network is used as the encoding structure, and the decoder of semantic segmentation model is used as the decoding structure.

4. The fully automatic railway centerline extraction method based on airborne point cloud as described in claim 3, characterized in that: The decoding structure contains four upsampling layers with a sampling interval of 2, which are used to restore the features generated by the encoding structure to the same size as the original input intensity image; After each upsampling layer, two convolutional layers are used to smooth the upsampling results, and the final output is a binary image of the railway centerline with the same size as the input image.

5. The fully automatic railway centerline extraction method based on airborne point cloud as described in claim 4, characterized in that: The two-dimensional intensity image is input into the trained convolutional neural network model RWNet to predict the binary image of the rail centerline and extract the two-dimensional rail centerline, including: The RWNet convolutional neural network model predicts binary images of the railway track centerline; The two-dimensional rail centerline is extracted by vectorizing the binary image of the rail centerline.

6. The fully automatic railway centerline extraction method based on airborne point cloud as described in claim 5, characterized in that: The railway centerline is stored as a node, including: After vectorizing the left and right rail lines, interpolate nodes at the same spacing. Save the lines as nodes, with a node spacing of S.

7. The fully automatic railway centerline extraction method based on airborne point cloud as described in claim 6, characterized in that: The current elevation of the railway track is a local elevation maximum.

8. The fully automatic railway centerline extraction method based on airborne point cloud as described in claim 7, characterized in that: Update node elevations to obtain the 3D track line. Extract the railway centerline based on the 3D track line, including: Obtain the 3D lines of the left and right railway tracks; For the i-th node on the three-dimensional line of the left rail Calculate each node on the 3D line of the right rail. its straight-line distance ; Find and straight-line distance The smallest right rail 3D line node ; according to and The railway centerline node was calculated. coordinate: =（ + ） / 2； For each node on the three-dimensional line of the left rail, repeat the calculation and connect them sequentially. This generates the railway centerline.

9. A fully automated railway centerline extraction system based on airborne point clouds, characterized in that: The system is used to implement the method according to any one of claims 1-8, comprising: The planar projection module is used to acquire point cloud data and obtain a two-dimensional intensity image after planar projection. The annotation module is used to annotate railway track lines in two-dimensional intensity images, using the two-dimensional intensity images and railway track lines as training data in pairs. The model training module is used to build a convolutional neural network model RWNet for extracting railway track lines from two-dimensional intensity images. It is trained by inputting training data to obtain the trained convolutional neural network model RWNet. The prediction module is used to input the two-dimensional intensity image into the trained convolutional neural network model RWNet, predict the binary image of the rail centerline, extract the two-dimensional rail centerline, and save the rail centerline in the form of nodes. The elevation extraction module is used to read point cloud data. For each node on the centerline of the railway, it searches for the current elevation of the current position of the railway in the point cloud data along a preset range perpendicular to the centerline of the railway. The railway centerline extraction module is used to update node elevations, obtain 3D lines of the railway track, and extract the railway centerline based on the 3D lines of the railway track.

Images