Method and system for detecting forward obstacle of unsupervised train

Abstract

The invention provides an unsupervised train forward obstacle detection method and system, and belongs to the technical field of image processing. A pre-trained detection model is used for processing a rail transit scene image collected by a train front-end camera, and detection of forward obstacles of a train is achieved; wherein in the detection model, a semantic segmentation unit is used for extracting all real track pixel points in an image, and a reconstruction unit is used for extracting track edge effective key points, fitting rule lines and reconstructing a track region with a complete rule on the basis of a semantic segmentation result according to geometric feature constraints of a track; the difference module is used for carrying out difference on the real track pixel points extracted through semantic segmentation and the reconstructed regular track area, locking the track missing part and obtaining obstacle pixel position information. According to the invention, deep learning and a traditional image processing algorithm are fused, and detection of various obstacles in the track area can be realized by using inherent geometric priori knowledge of the track.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an unsupervised train forward obstacle detection method and system. BACKGROUND

[0002] In the rail transit scene, timely monitoring of the train operating environment is of great significance to ensure the safe operation of the train. Train forward obstacle detection has become a key link to ensure train safety. With the development of computer vision technology, intelligent obstacle detection methods based on visual perception have become an important means of railway safety protection. Through intelligent analysis of the real-time image data of the railway environment collected by the camera at the front of the train, accurate detection of the forward obstacle can be achieved. Most of the mainstream train forward obstacle detection methods first use a semantic segmentation algorithm to extract the track, then use target detection to detect known common obstacles, and further determine the position of the obstacle. Although these methods have made important breakthroughs in dealing with key technical difficulties such as adverse weather interference such as rain, snow and fog, and long-distance small target detection, their detection capabilities are still limited to known obstacle categories, and there is a lack of effective detection of unknown obstacle categories that may appear in the track environment.

[0003] In view of this technical bottleneck, in recent years, some scholars have proposed an innovative solution based on image generation and difference analysis. The core idea of this method is to use deep learning models such as generative adversarial networks or variational autoencoders to construct a no-obstacle benchmark image library for the track scene, and to realize the detection of all categories of obstacles through pixel-level difference analysis of real-time image and generated image. However, there are two significant defects in existing research: first, most methods directly perform end-to-end generation modeling in the RGB three-channel image space, ignoring the structured features specific to the track scene, resulting in difficulty in ensuring the quality and stability of the generated image; second, difference analysis based on global images will introduce a lot of background noise and lighting change interference, causing serious false detection problems. Therefore, developing a robust obstacle detection algorithm with scene adaptability, and establishing a technical system that takes into account the comprehensiveness and accuracy of detection, has become a key scientific problem that needs to be overcome in the current field of intelligent safety protection for rail transit. SUMMARY

[0004] The present application aims to provide an unsupervised train forward obstacle detection method and system, which uses a semantic segmentation network to accurately extract the track pixel distribution in the image, then constructs a complete track based on the geometric prior of the track structure, and finally accurately locates the track damage part through a difference analysis algorithm, thereby realizing the detection of all types of obstacles in the track area, to solve at least one of the technical problems existing in the above background technology.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a method for unsupervised forward obstacle detection of a train, comprising:

[0007] Obtaining a rail transit scene image;

[0008] Processing the rail transit scene image collected by the front camera of the train using a pre-trained detection model to realize detection of the forward obstacle of the train; wherein the detection model comprises a semantic segmentation unit, a reconstruction unit and a difference unit; the semantic segmentation unit is used to extract all real track pixel points in the image; the reconstruction unit is used to extract effective key points of the track edge, fit regular lines and reconstruct a regular and complete track area on the basis of the semantic segmentation result according to the geometric feature constraint of the track; the difference module is used to perform difference between the real track pixel points extracted by the semantic segmentation and the reconstructed regular track area, lock the missing part of the track and obtain the pixel position information of the obstacle, so as to realize detection of the forward obstacle of the train.

[0009] As a further limitation of the first aspect of the present application, collect the rail transit scene image data in front of the train, label the actual track pixel area in the image and make a semantic segmentation data set; use the track data set to train an advanced semantic segmentation unit; use the trained semantic segmentation unit to extract real track pixel points in the image and output a binary semantic segmentation result, wherein the pixel value of the recognized track area is 255 and the pixel value of the non-track area is uniformly set to 0.

[0010] As a further limitation of the first aspect of the present application, use image filtering to process the semantic segmentation result and extract real track edge information; set multiple groups of parallel scanning lines, extract the intersection points of the scanning lines and the track edge as the track edge key points.

[0011] As a further limitation of the first aspect of the present application, for adjacent edge key points, set a slope judgment condition according to the geometric characteristics of the track, exclude abnormal points extracted due to track loss and retain the final effective track edge key points; fit the effective key points as a regular track edge; perform pixel filling inside the track edge to generate a regular and complete binary track map.

[0012] As a further limitation of the first aspect of the present application, calculate the difference value of the pixel points at the same position in the real track result map and the reconstructed regular and complete track map; based on the binary result map obtained by difference processing, extract the contour edge information of all pixel values of 255 and superimpose and draw these edge contours on the original input image to realize visual presentation of the obstacle position.

[0013] As a further limitation of the first aspect of the present application, the difference value of the pixel points at the same position in the real track result map and the reconstructed regular and complete track map is:

[0014]

[0015] wherein, I t (x, y) is the pixel value of the (x, y) position in the real track result image extracted by semantic segmentation, B t (x, y) is the pixel value of the (x, y) position in the reconstructed complete regular track image, T is a difference threshold value; when the difference between the two is greater than the set threshold value, the pixel value of the new image generated in this area is 255, otherwise it is 0.

[0016] In a second aspect, the present application provides an unsupervised train forward obstacle detection system, comprising:

[0017] An acquisition module is configured to acquire a rail transit scene image.

[0018] A processing module is configured to process the rail transit scene image collected by the front camera of the train by using a pre-trained detection model, so as to realize detection of the train forward obstacle. The detection model comprises a semantic segmentation unit, a reconstruction unit and a difference unit. The semantic segmentation unit is configured to extract all real track pixel points in the image. The reconstruction unit is configured to extract track edge effective key points, fit regular lines and reconstruct a regular complete track area on the basis of the semantic segmentation result, according to the geometric feature constraint of the track. The difference module is configured to perform difference between the real track pixel points extracted by semantic segmentation and the reconstructed regular track area, lock the missing part of the track and obtain obstacle pixel position information, so as to realize detection of the train forward obstacle.

[0019] In a third aspect, the present application provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implement the unsupervised train forward obstacle detection method of the first aspect.

[0020] In a fourth aspect, the present application provides a computer device comprising a memory and a processor, wherein the processor and the memory are in communication with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the unsupervised train forward obstacle detection method of the first aspect.

[0021] In a fifth aspect, the present application provides an electronic device comprising a processor, a memory and a computer program, wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to make the electronic device execute the instructions for implementing the unsupervised train forward obstacle detection method of the first aspect.

[0022] The present application has the advantages that: the deep learning technology and the traditional image processing algorithm are fused, and through effective utilization of the inherent geometric prior knowledge of the track, various obstacles in the track area can be detected.

[0023] The advantages of the additional aspects of the present application will become more apparent in the description below, or will be understood by practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0025] Figure 1 The train forward obstacle detection method flowchart described in the embodiments of the present application.

[0026] Figure 2 The experimental results of the train forward obstacle detection method described in the embodiments of the present application on the train forward obstacle data are shown in the figure. Among them, Figure 2 (a) is the original input image, Figure 2 (b) is the track pixel result image extracted by semantic segmentation, Figure 2 (c) is the reconstructed regular track image, Figure 2 (d) is the obstacle pixel information image obtained by difference, Figure 2 (e) is the final obstacle detection result image. DETAILED DESCRIPTION

[0027] The embodiments of the present application will be described in detail below, and the examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0028] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the field to which the present application belongs.

[0029] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as such.

[0030] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprising," "including," "containing," and "having" and the like, when used in the specification, specify the presence of stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0031] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. Those skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0032] In order to facilitate the understanding of the present application, the present application will be further explained and described in specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present application.

[0033] Those skilled in the art should understand that the drawings are only schematic views of the embodiments, and the components in the drawings are not necessarily essential for the implementation of the present application.

[0034] The present application is directed to the problem of difficult identification of unknown obstacles in the forward scene of the current rail transit train. In combination with the geometric characteristics of the track, a kind of unsupervised train forward obstacle detection method is designed by using the principle of image difference. The method adopts a three-stage progressive detection framework: first, the real track pixel points existing in the image are extracted using a semantic segmentation method; second, based on the segmentation result, the key points of the track edge are extracted according to the prior rule track geometric characteristics, and the rule track is reconstructed; finally, based on the principle of image difference, the semantic segmentation result is differentiated with the reconstructed rule track result, and the identification of the obstacles in the track area is realized. The method innovatively combines deep learning technology and traditional image processing algorithm, and can realize the detection of various obstacles in the track area by effectively utilizing the inherent geometric prior knowledge of the track.

[0035] Embodiment 1

[0036] In this embodiment 1, first, an unsupervised train forward obstacle detection system is provided, comprising: an acquisition module for acquiring a rail transit scene image; a processing module for processing the rail transit scene image collected by the train front camera using a pre-trained detection model to realize the detection of the train forward obstacles.

[0037] In this embodiment, the above system is used to realize an unsupervised train forward obstacle detection method, which includes: using an acquisition module to acquire a rail transit scene image; for example, a train front camera is used as the acquisition module to collect the scene image. A processing module is used to process the rail transit scene image collected by the train front camera using a pre-trained detection model to realize detection of the train forward obstacle. The detection model includes a semantic segmentation unit, a reconstruction unit and a difference unit; the semantic segmentation unit is used to extract all real track pixel points in the image; the reconstruction unit is used to extract track edge effective key points based on the semantic segmentation result, fit regular lines, and reconstruct a regular complete track area according to the geometric feature constraint of the track; the difference module is used to perform difference between the real track pixel points extracted by the semantic segmentation and the reconstructed regular track area, lock the track missing part, and obtain the obstacle pixel position information, so as to realize detection of the train forward obstacle.

[0038] The rail transit scene image data in front of the train is collected, the actual track pixel area in the image is labeled, and a semantic segmentation data set is made; an advanced semantic segmentation unit is trained using the track data set; the real track pixel points in the image are extracted using the trained semantic segmentation unit, and a binary semantic segmentation result is output, wherein the pixel value of the recognized track area is 255, and the pixel value of the non-track area is uniformly set to 0. The semantic segmentation result is processed using image filtering to extract the real track edge information; a plurality of parallel scanning lines are set, the intersection points of the scanning lines and the track edge are extracted as the track edge key points. For adjacent edge key points, according to the geometric characteristics of the track, a slope judgment condition is set to exclude abnormal points extracted due to track missing, and the final effective track edge key points are retained; the effective key points are fitted as regular track edges; the inside of the track edge is filled with pixels to generate a regular complete binary track graph. The difference value of the pixel points at the same position in the real track result graph and the reconstructed regular complete track graph is calculated; based on the binary result graph obtained by the difference processing, the contour edge information of all pixel values of 255 is extracted, and these edge contours are superimposed and drawn on the original input image to realize visual presentation of the obstacle position. The difference value of the pixel points at the same position in the real track result graph and the reconstructed regular complete track graph is calculated as:

[0039]

[0040] wherein, I t (x, y) is the pixel value of the (x, y) position in the real track result graph extracted by the semantic segmentation, B t(x, y) is the pixel value of the (x, y) position in the reconstructed complete rule track image, T is the difference threshold value; when the difference is greater than the set threshold value, the pixel value of the new image generated by the region is 255, otherwise it is 0.

[0041] Embodiment 2

[0042] In this embodiment 2, the specific implementation process of the provided unsupervised train front obstacle detection method includes the following steps:

[0043] Step S1, using an advanced semantic segmentation algorithm to process the track scene image collected by the train front camera, extracting all the real track pixel points in the image.

[0044] The step S1 specifically includes the following steps:

[0045] Step 1.1, collect the track scene image data in front of the train, label the actual track pixel area in the image, and make a semantic segmentation data set;

[0046] Step 1.2, using the track data set to train an advanced semantic segmentation model;

[0047] Step 1.3, using the trained semantic segmentation model to extract the real track pixel points in the image, and output the binary semantic segmentation result, wherein the pixel value of the recognized effective track area is 255, and the pixel value of the non-track area is uniformly set to 0.

[0048] Step S2: based on the geometric feature constraint of the track, extracting the track edge effective key points based on the semantic segmentation result, fitting the rule line, and reconstructing the rule complete track area.

[0049] The step S2 specifically includes the following steps:

[0050] Step 2.1, using a traditional image filtering algorithm to process the semantic segmentation result, extracting the real track edge information;

[0051] Step 2.2, set multiple groups of parallel scanning lines, extract the intersection points of the scanning lines and the track edges as the track edge key points;

[0052] Step 2.3, for adjacent edge key points, according to the geometric characteristics of the track, set the slope judgment condition, exclude the abnormal points extracted due to track loss, and retain the final effective track edge key points;

[0053] Step 2.4, line fitting of effective key points as rule track edge;

[0054] Step 2.5, pixel filling inside the track edge to generate a rule complete binary track image.

[0055] Step S3, difference between the real track pixel points extracted by semantic segmentation and the reconstructed regular track area, lock the missing part of the track, obtain the obstacle pixel position information, so as to realize the detection of the forward obstacle of the train.

[0056] The step S3 specifically includes the following steps:

[0057] Step 3.1, calculate the difference value of the pixel points at the same position in the real track result image and the reconstructed regular complete track image, and the formula is as follows:

[0058]

[0059] Wherein, I t (x,y) is the pixel value of (x,y) position in the real track result image extracted by semantic segmentation, B t (x,y) is the pixel value of (x,y) position in the reconstructed complete regular track image, and T is the difference threshold value. When the difference value is greater than the set threshold value, the new image pixel value generated in this area is 255, otherwise it is 0.

[0060] Step 3.2, based on the binary result image obtained by difference processing, extract all the edge information of the pixel value of 255, and superimpose and draw these edge contours on the original input image to realize the visual presentation of the obstacle position.

[0061] As described above, the method of the embodiment first extracts the real track pixel points existing in the image using the semantic segmentation method, then reconstructs the track according to the geometric constraints of the regular track, and finally, according to the image difference principle, the original segmentation result is different from the reconstructed track, so as to realize the detection of all obstacles in the track. Compared with the existing train forward obstacle detection method, this method combines the powerful pixel-level feature extraction ability of deep learning and the prior knowledge of track structure, which can not only accurately identify various known obstacles in the track area, but more importantly, it breaks through the limitations of traditional methods in unknown category obstacle detection, and realizes the reliable detection of all obstacles in the track area.

[0062] Embodiment 3

[0063] The embodiment 3 provides a non-transitory computer readable storage medium for storing computer instructions, which are executed by a processor to realize the unsupervised train forward obstacle detection method as described above, which includes:

[0064] Obtain the track traffic scene image;

[0065] The track transportation scene image collected by the front camera of the train is processed by using a pre-trained detection model to realize detection of the forward obstacle of the train; wherein the detection model comprises a semantic segmentation unit, a reconstruction unit and a difference unit; the semantic segmentation unit is used to extract all real track pixel points in the image; the reconstruction unit is used to extract effective key points of the track edge, fit regular lines and reconstruct a regular complete track area on the basis of the semantic segmentation result according to the geometric feature constraint of the track; the difference module is used to perform difference between the real track pixel points extracted by the semantic segmentation and the reconstructed regular track area, lock the missing part of the track and obtain the pixel position information of the obstacle, so as to realize detection of the forward obstacle of the train.

[0066] Embodiment 4

[0067] The embodiment 4 provides a computer device, comprising a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the unsupervised train forward obstacle detection method as described above, which comprises:

[0068] Obtaining a track transportation scene image;

[0069] The track transportation scene image collected by the front camera of the train is processed by using a pre-trained detection model to realize detection of the forward obstacle of the train; wherein the detection model comprises a semantic segmentation unit, a reconstruction unit and a difference unit; the semantic segmentation unit is used to extract all real track pixel points in the image; the reconstruction unit is used to extract effective key points of the track edge, fit regular lines and reconstruct a regular complete track area on the basis of the semantic segmentation result according to the geometric feature constraint of the track; the difference module is used to perform difference between the real track pixel points extracted by the semantic segmentation and the reconstructed regular track area, lock the missing part of the track and obtain the pixel position information of the obstacle, so as to realize detection of the forward obstacle of the train.

[0070] Embodiment 5

[0071] The embodiment 5 provides an electronic device, comprising a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory to make the electronic device execute the instructions of the unsupervised train forward obstacle detection method as described above, which comprises:

[0072] Obtaining a track transportation scene image;

[0073] The pre-trained detection model is used to process the track transportation scene image collected by the front camera of the train, so as to realize the detection of the forward obstacle of the train; wherein the detection model comprises a semantic segmentation unit, a reconstruction unit and a difference unit; the semantic segmentation unit is used to extract all real track pixels in the image; the reconstruction unit is used to extract effective key points of the track edge, fit a regular line, and reconstruct a regular complete track area on the basis of the semantic segmentation result according to the geometric feature constraint of the track; the difference module is used to perform difference between the real track pixels extracted by the semantic segmentation and the reconstructed regular track area, lock the missing part of the track, and obtain the pixel position information of the obstacle, so as to realize the detection of the forward obstacle of the train.

[0074] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows 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 apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0076] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0077] These computer program instructions can also be loaded into a computer or other programmable data processing devices, to cause a series of operational steps to be performed on the computer or other programmable data processing devices, so as to generate a computer implemented process, thus the instructions executed on the computer or other programmable data processing devices provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one flow or multiple flows and / or the functions specified in the block

[0078] The above describes the specific embodiments of the present application in conjunction with the drawings, but is not a limitation on the scope of protection of the present application, those skilled in the art should understand that on the basis of the technical solutions disclosed in the present application, various modifications or changes made by those skilled in the art without the need for creative labor should be covered within the scope of protection of the present application.

Claims

1. An unsupervised train forward obstacle detection method, characterized in that: include: Acquire rail transit scene images; A pre-trained detection model is used to process rail transit scene images captured by a train's front-end camera to detect obstacles ahead of the train. The detection model includes a semantic segmentation unit, a reconstruction unit, and a difference unit. The semantic segmentation unit is used to extract all real track pixels in the image. The reconstruction unit is used to extract valid key points on the track edge based on the semantic segmentation results, fit regular lines, and reconstruct a regular and complete track area according to the geometric feature constraints of the track. The difference module is used to differentiate the real track pixel points extracted by semantic segmentation from the reconstructed regular track area, lock the missing part of the track, obtain the obstacle pixel position information, and thus realize the detection of obstacles ahead of the train.

2. The unsupervised train forward obstacle detection method according to claim 1, characterized in that: Collect rail transit scene image data from the train's forward direction, annotate the actual track pixel areas in the image, and create a semantic segmentation dataset. Use the track dataset to train an advanced semantic segmentation unit. Use the trained semantic segmentation unit to extract the actual track pixel points in the image and output a binary semantic segmentation result, where the pixel values ​​of the identified track area are 255 and the pixel values ​​of the non-track area are uniformly set to 0.

3. The unsupervised train forward obstacle detection method according to claim 1, characterized in that: Image filtering is used to process the semantic segmentation results to extract the real track edge information; multiple sets of parallel scan lines are set, and the intersection points between the scan lines and the track edges are extracted as track edge key points.

4. The unsupervised train forward obstacle detection method according to claim 3, characterized in that: For adjacent edge key points, the slope judgment condition is set according to the track geometric characteristics, the abnormal points extracted due to track missing are excluded, and the final valid track edge key points are retained; the valid key points are fitted with lines as regular track edges; the pixels inside the track edges are filled to generate a regular and complete binary track map.

5. The unsupervised train forward obstacle detection method according to claim 1, characterized in that: The difference between the pixel at the same position in the real track result image and the reconstructed regular complete track image is calculated; based on the binary result image obtained by the difference processing, the contour edge information of all pixel values ​​​​of 255 is extracted, and these edge contours are superimposed and drawn on the original input image to achieve a visual presentation of the obstacle position.

6. The unsupervised train forward obstacle detection method according to claim 5, characterized in that: The difference between the pixel at the same position in the real orbit result map and the reconstructed regular complete orbit map is: Among them, I t (x, y) is the pixel value at position (x, y) in the real track result map extracted by semantic segmentation, B t (x, y) is the pixel value at the (x, y) position in the reconstructed complete regular track image, and T is the difference threshold. When the difference between the two is greater than the set threshold, the pixel value of the new image generated in this area is 255, otherwise it is 0.

7. An unsupervised train forward obstacle detection system, characterized in that: include: An acquisition module is used to acquire rail transit scene images; A processing module is used to process rail transit scene images captured by the train's front-end camera using a pre-trained detection model to detect obstacles ahead of the train. The detection model includes a semantic segmentation unit, a reconstruction unit, and a difference unit. The semantic segmentation unit is used to extract all real track pixels in the image. The reconstruction unit is used to extract valid key points on the track edge based on the semantic segmentation results, fit regular lines, and reconstruct a regular and complete track area according to the geometric feature constraints of the track. The difference module is used to differentiate the real track pixel points extracted by semantic segmentation from the reconstructed regular track area, lock the missing part of the track, obtain the obstacle pixel position information, and thus realize the detection of obstacles ahead of the train.

8. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, the unsupervised train forward obstacle detection method according to any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the unsupervised train forward obstacle detection method according to any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing the unsupervised train forward obstacle detection method according to any one of claims 1 to 6.

Images