Abnormality detection method for ballast bed cover plate of railway track

Abstract

The invention relates to the technical field of image processing, in particular to an anomaly detection method for a ballast bed cover plate of a railway track. Comprising the following steps: synchronously acquiring track area images from different angles by using at least two cameras, calculating a depth value of the ballast bed cover plate according to parallax of the same target track area in imaging of different cameras, and constructing a three-dimensional model of the ballast bed cover plate; in the high-speed running process of the train, real-time compensation processing is carried out on the image; a periodic action sequence including different illumination intensities, focal length adjustment and shooting frame rate changes is set, the duration of each action is set according to the train running speed and the camera shooting frequency, and multiple sets of track images are collected during the duration of each action. According to the railway track ballast bed cover plate anomaly detection method, the depth value of the ballast bed cover plate is obtained, the tilting anomaly of the cover plate is recognized, real-time compensation processing is carried out on the image, and the detection precision and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method for detecting abnormalities of the roadbed cover plate of railway tracks. Background Art

[0002] With the rapid development of railway transportation, the safe operation of railway tracks is of crucial importance. As an important part of railway tracks, the state of the roadbed cover plate directly affects the safety and stability of train operation. During the high-speed running of trains, the roadbed cover plate is prone to abnormal conditions such as missing and warping. If not detected and repaired in time, it is extremely likely to cause safety accidents.

[0003] Currently, the detection of roadbed cover plates mainly relies on manual inspections and traditional detection methods. Manual inspections are inefficient, labor-intensive, and greatly affected by human factors, making it difficult to ensure the accuracy and timeliness of detection. Traditional detection methods, such as visual detection based on a single camera, since a single camera cannot obtain depth information, can only observe some obvious changes on the surface of the cover plate, and it is difficult to accurately detect subtle abnormalities with depth changes such as warping. At the same time, during the high-speed running of trains, the images are prone to problems such as blurring and deformation, further affecting the detection accuracy.

[0004] For example, in the patent with Chinese Patent Publication No. CN119624945A and the name "Track Geometry Detection Line Feature Region Recognition Method, System and Electronic Device", it includes: determining the track line feature image to be recognized; inputting the track line feature image to be recognized into the line feature region recognition model to obtain the line feature region recognition result output by the line feature region recognition model; wherein, the line feature region recognition model is obtained after being trained based on the sample data of the line feature image and the regional image annotation of the sample data; the line feature region recognition model is used to perform multiple feature judgments on the fused data after the space-time synchronization of the track line feature image to be recognized and the track geometry detection waveform data to obtain the line feature region recognition result. The disadvantage is that it can only observe some obvious changes on the surface of the cover plate, and it is difficult to accurately detect subtle abnormalities with depth changes such as warping. Summary of the Invention

[0005] Aiming at the problem that the abnormal detection of railway tracks in the prior art cannot recognize the abnormal warping of the cover plate, the present invention provides a method for detecting abnormalities of the roadbed cover plate of railway tracks, which obtains the depth information of the roadbed cover plate, recognizes the abnormal warping of the cover plate, and performs real-time compensation processing on the image to improve the detection accuracy and efficiency.

[0006] To achieve the above technical purpose, a technical solution provided by the present invention is a method for detecting abnormalities of the roadbed cover plate of railway tracks, including the following steps: S1. Use at least two cameras to synchronously collect images of the track area from different angles. Calculate the depth value of the track bed cover plate according to the parallax of the same target track area in the images captured by different cameras, and construct a three-dimensional model of the track bed cover plate. S2. During the high-speed train operation, perform real-time compensation processing on the images. S3. Set a periodic action sequence including different light intensities, focal length adjustments, and shooting frame rate changes. Set the duration of each action according to the train running speed and the camera shooting frequency, and collect multiple groups of track images during the duration of each action. S4. Extract the feature points of the images under each action, perform feature matching and depth value fusion, and determine whether the abnormality is caused by the track bed cover plate body, camera shooting parameters, or train running speed changes. S5. Based on the judgment of the abnormal results, construct a residual image law model, perform real-time correction on the subsequent images, generate candidate regions of the abnormality through the region proposal network, classify and perform position regression on the candidate regions, and determine whether there are abnormalities such as missing or warping of the track bed cover plate and locate them.

[0007] In this technical solution, by using at least two cameras to synchronously collect images of the track area from different angles and simulating the binocular vision principle, the depth value of the track bed cover plate can be accurately calculated, and a fine three-dimensional model can be constructed, providing rich three-dimensional space data for subsequent abnormality detection and helping to more comprehensively understand the state of the track bed cover plate. During the high-speed train operation, real-time compensation processing is performed on the images, effectively eliminating motion blur and enhancing contrast, ensuring that clear and high-quality images can be obtained even in a high-speed motion environment. Setting a periodic action sequence including different light intensities, focal length adjustments, and shooting frame rate changes, and setting the duration of each action according to the train running speed and the camera shooting frequency, enables the detection method to adapt to different environmental conditions, improving the robustness and applicability of the detection. Extracting the feature points of the images under each action, performing feature matching and depth value fusion, can accurately determine whether the abnormality is caused by the track bed cover plate body, camera shooting parameters, or train running speed changes, comprehensively considering various factors, and improving the accuracy and comprehensiveness of abnormality judgment. Based on the judgment of the abnormal results, construct a residual image law model, and use a deep learning network to perform real-time correction and analysis on the subsequent images, generate candidate regions of the abnormality through the region proposal network, and perform classification and position regression, which can achieve accurate positioning and classification of abnormalities such as missing and warping of the track bed cover plate, not only improving the accuracy of detection, but also significantly enhancing the detection efficiency.

[0008] The present invention is further set as follows: The method further includes camera setting optimization, and the camera setting optimization is as follows: Set a camera to take pictures at three different positions. Select the positions according to the layout of the track bed cover plate and the candidate areas of anomalies. Control the camera to switch between different positions through a mechanical device. After each switch, use a calibration device to calibrate the internal and external parameters of the camera by taking pictures of a calibration plate with a known size. Compare the differences in the images at different positions, use an image stitching algorithm to stitch the images, and analyze the differences in the images at different positions in the stitched image to assist in judging the cause of the anomaly; Or, set three cameras at the same position to take pictures from different angles or with different parameters; when taking pictures from different angles, select the angles to ensure obtaining different side information of the cover plate in the track area. When taking pictures with different parameters, the parameters include exposure time, focal length, and aperture. Use automatic control and synchronous shooting technology to ensure that the time bases of the images are the same. When comparing the images, use image registration and fusion technology to decompose the images into sub-bands of different scales. According to the characteristics of each scale sub-band of the images from different cameras, use the weighted average method to fuse them, establish a correspondence relationship model between the image differences and the anomaly factors, and determine the cause of the anomaly based on the difference points.

[0009] In this technical solution, a mechanical device is used to control a single camera to switch between three different positions, forming a multi-position camera switching strategy to ensure that the camera can comprehensively cover all areas of the track bed cover plate, especially the areas prone to anomalies. After each switch, a calibration device is used to take pictures of a calibration plate with a known size to accurately calibrate the internal and external parameters of the camera, thus ensuring that the images taken at different positions have a consistent geometric relationship, laying a solid foundation for subsequent image stitching and difference analysis. An image stitching algorithm is used to seamlessly stitch the images taken at different positions into a complete image, facilitating the inspectors to comprehensively observe the state of the track bed cover plate. At the same time, by analyzing the differences in the images at different positions in the stitched image, it is possible to more accurately assist in judging the cause of the anomaly, improving the accuracy and efficiency of the detection. Set three cameras at the same position to take pictures from different angles or with different parameters (such as exposure time, focal length, aperture, etc.), forming a multi-camera array strategy to obtain more comprehensive and diverse information about the track bed cover plate. The multi-angle and multi-parameter shooting methods help to capture the subtle changes of the track bed cover plate, providing more clues for anomaly detection. Image registration and fusion technology are used to accurately align and fuse the images taken by different cameras. The fused image not only retains the useful information taken by each camera but also fuses the characteristics of different scale sub-bands through the weighted average method, further enriching the image features and improving the accuracy of anomaly detection. Establish a correspondence relationship model between the image differences and the anomaly factors, and determine the cause of the anomaly based on the difference points in the fused image, which can comprehensively consider various factors such as light changes and shooting angle differences, so as to more accurately judge whether the anomaly stems from the track bed cover plate itself or other external factors.

[0010] The present invention is further configured such that in step S1, calculating the depth value of the track bed cover plate according to the parallax in the imaging of different cameras in the same target track area and constructing the three-dimensional model of the track bed cover plate includes: dividing the image obtained by one of the cameras into multiple small image blocks, searching for the most similar area in the image of the other camera, obtaining the pixel point parallax by calculating the displacement of the image blocks, and calculating the depth value according to the principle of triangulation to construct the three-dimensional model of the track bed cover plate.

[0011] In this technical solution, the image obtained by one of the cameras is divided into multiple small image blocks, and the most similar area is searched for in the image of the other camera. By calculating the displacement of the image blocks, the pixel point parallax is obtained, and then the depth value of the track bed cover plate is accurately calculated according to the principle of triangulation. This method based on the stereo matching algorithm can make full use of the principle of binocular vision to improve the accuracy and reliability of the depth value calculation. Using the calculated depth value, a three-dimensional model of the track bed cover plate can be constructed, which can intuitively display the shape, position and spatial relationship of the track bed cover plate, providing rich three-dimensional space data for subsequent anomaly detection. Through the three-dimensional model, the inspectors can more comprehensively understand the state of the track bed cover plate, which helps to discover potential anomalies or defects. Based on the accurately calculated depth value and the constructed three-dimensional model, fine detection of the track bed cover plate can be realized. By comparing the three-dimensional model with the model in the normal state, it is easier to discover the anomalies or deformations of the track bed cover plate, which not only improves the accuracy of anomaly detection, but also helps to discover potential safety hazards in advance. The method of simulating the principle of binocular vision by the stereo matching algorithm has strong robustness and can work stably under different lighting conditions, shooting angles and distances. This robustness enables the method to adapt to various complex railway track environments and improves the stability and reliability of the system.

[0012] The present invention is further configured such that in step S2, the real-time compensation processing of the image during the high-speed running of the train includes: Adopting an algorithm based on the motion blur model, calculating the degree of image motion blur according to the train running speed and the camera shooting parameters, and restoring the blurred image by the inverse filtering method; Extracting feature points from consecutive frame images through the feature point detection algorithm, calculating the motion vectors between adjacent frames of the feature points according to the optical flow method to obtain the overall motion information of the image, reversely compensating the blurred image based on the overall motion information of the image, resampling by the bilinear interpolation algorithm, and processing the image in regions according to the adaptive histogram equalization algorithm.

[0013] In this technical solution, the algorithm based on the motion blur model, combined with the train speed and camera shooting parameters, can accurately calculate the degree of motion blur of the image. Through the inverse filtering method, the blurred image is restored, which effectively eliminates the image blur caused by the high-speed movement of the train, making the image edge clearer and the details richer. The feature point detection algorithm is used to extract feature points in continuous frame images, and the motion vector between adjacent frames of the feature points is calculated by the optical flow method to obtain the overall motion information of the image. Based on this information, the blurred image is reversely compensated to correct the image distortion caused by motion blur. Subsequently, the bilinear interpolation algorithm is used to resample the image, further smooth the image, eliminate the jagged or distortion phenomenon generated during the compensation process, and significantly improve the image quality. The adaptive histogram equalization algorithm is used to process the image in different regions. This algorithm can automatically adjust the contrast according to the brightness distribution of different regions of the image, making the dark details in the image clearer and the bright details richer. At the same time, the regional processing method can avoid the noise amplification and detail loss problems caused by global histogram equalization, so that the image contrast can be improved more naturally and evenly. By enhancing the contrast of the image, the texture, edge and other features of the roadbed cover are more prominent, which helps the subsequent anomaly detection algorithm to more accurately identify abnormal phenomena such as missing and warping of the roadbed cover. At the same time, the improvement of contrast also enables the inspection personnel to observe the image more clearly, improving the efficiency and accuracy of the inspection.

[0014] The present invention is further configured as follows: in step S3, the feature points of the image under each action are extracted, and the feature matching and depth value fusion are performed, including: using an image feature extraction and matching algorithm to process images taken under different actions, extracting feature points and calculating descriptors between feature points, and matching feature points by comparing descriptor similarities; and analyzing the differences in feature points between images of different actions in combination with depth values.

[0015] In this technical solution, by using the image feature extraction and matching algorithm, significant feature points in the captured images under different actions can be accurately extracted, and the descriptors between these feature points are calculated. By comparing the similarity of the descriptors, the feature points in different images can be accurately matched, providing a reliable basis for subsequent depth value fusion and anomaly judgment. Combining with the depth value, the differences of feature points between different action images are analyzed. The depth value provides the three-dimensional coordinates of the feature points, which helps to understand the position and relationship of the feature points in space. By analyzing the feature point differences, it can be determined whether these differences are caused by changes in the ballast cover body, such as anomalies like missing or warping, or by factors such as camera shooting parameters and train driving speed changes. This depth value fusion analysis method improves the accuracy and comprehensiveness of anomaly judgment. Based on the results of feature matching and depth value fusion, it can be accurately determined whether there is an anomaly in the ballast cover and locate the position of the anomaly, which not only improves the accuracy of anomaly detection but also helps to take timely measures for repair, thus ensuring the safe operation of the railway track. The feature extraction and matching algorithm has high computational efficiency and can quickly process a large amount of image data. At the same time, by analyzing in combination with the depth value, the detection method has stronger robustness to factors such as different lighting conditions, shooting angles, and distances, and can work stably in various complex railway track environments.

[0016] The present invention is further set as follows: In step S4, the determination that the anomaly is due to changes in the ballast cover body, camera shooting parameters, or train driving speed includes: By comparing the changes in the matching values and depth values of feature points under different action sequences, if the feature point matching fails or there are local depth mutations or surface unevenness in the depth value, it is determined that the ballast cover body is abnormal; Analyze the influence of different exposure times, focal lengths, and apertures on image features. If the feature points are successfully matched but there is an overall depth offset or image blurring after depth value fusion, it is determined that the camera shooting parameters are abnormal; According to the train driving speed and the camera shooting frequency, set the duration of the action sequence. If there is motion blurring or image misalignment in the results of feature point matching and depth value fusion, it is determined that the anomaly is caused by the change in the train driving speed.

[0017] In this technical solution, when the feature point matching fails or the depth value has a local depth mutation or the surface is uneven, it is determined that the roadbed cover body is abnormal, and the physical changes of the roadbed cover, such as missing and warping, can be directly detected, providing accurate information for timely measures to repair. When the feature point matching is successful but the overall depth offset and image blur appear after the depth value fusion, it is determined that the camera shooting parameters are abnormal, and the image quality problems caused by improper camera parameter settings (such as exposure time, focal length, aperture, etc.) can be identified, which helps to adjust the camera parameters to obtain higher quality images. The duration of the action sequence is set according to the train speed and the camera shooting frequency. When the feature point matching and depth value fusion results show motion blur and image dislocation, it is determined that the abnormality is caused by the change in train speed. The impact of the change in train speed on the image acquisition and detection process can be detected, which helps to optimize the detection algorithm to adapt to different train speeds. By comprehensively considering multiple factors to determine the source of the abnormality, misjudgment and missed judgment caused by single factor judgment are avoided, which not only improves the accuracy of abnormal judgment, but also makes the detection method more comprehensive and reliable.

[0018] The present invention is further configured as follows: in step S5, the step of constructing the afterimage regularity model based on the abnormal result judgment comprises: The BakeMesh method is used to create the afterimage effect by duplicating the geometric model of the roadbed cover and adjusting its transparency and position at different time points; The collected images of the roadbed cover are post-processed using screen post-processing technology to generate afterimage effects. The movement trajectory and state changes of the roadbed cover are analyzed by comparing the afterimage images at different time points. The vertex positions of the geometric model of the roadbed cover are adjusted through the vertex offset method to simulate its state changes at different time points and generate a continuous afterimage effect.

[0019] In this technical solution, the BakeMesh method is adopted. By copying the geometric model of the track bed cover plate and adjusting its transparency and position at different time points to create an afterimage effect, it can intuitively display the state changes of the track bed cover plate at different time points, providing visual support for analysts. Analysts can more clearly observe the movement trajectory and state change trend of the track bed cover plate. Using screen post-processing technology to post-process the collected images of the track bed cover plate to generate an afterimage effect can further enhance the fidelity of the afterimage effect, making the afterimage image closer to the real situation and providing more accurate visual information for analysts. By using the vertex offset method to adjust the vertex positions of the geometric model of the track bed cover plate and simulate its state changes at different time points to generate a continuous afterimage effect, it can more precisely control the generation process of the afterimage effect, making the afterimage effect more continuous and natural and providing a smoother viewing experience for analysts. By comparing the afterimage images at different time points, the movement trajectory and state changes of the track bed cover plate can be analyzed, which helps to understand the movement law and state change trend of the track bed cover plate at different time points and provides important data support for subsequent anomaly detection and state assessment. At the same time, by analyzing the afterimage law, the future state changes of the track bed cover plate can also be predicted, providing a basis for taking preventive measures for repair in advance.

[0020] The present invention is further configured as follows: in step S5, the real-time correction of subsequent images includes: Performing denoising and contrast enhancement preprocessing on the collected images of the track bed cover plate according to a convolutional neural network; Integrating a multi-head self-attention module into the deep learning network to capture complex features in the images of the track bed cover plate; Designing a cross-layer weighted cascade structure to fuse deep network insights and shallow information and optimize the regression convergence of defect boundaries.

[0021] In this technical solution, the collected ballast cover plate images are preprocessed by a convolutional neural network for denoising and enhancing contrast. The noise in the images is effectively eliminated, and the clarity and contrast of the images are improved, making subsequent feature extraction and defect detection more accurate, and providing higher-quality image data for the entire detection process. The multi-head self-attention module is integrated into the deep learning network, enabling the network to capture complex features in the ballast cover plate images. The self-attention mechanism can focus on the key information in the images, while the multi-head design can capture features from different aspects, improving the diversity and robustness of feature extraction, helping to comprehensively understand the state of the ballast cover plate, and improving the accuracy of anomaly detection. A cross-layer weighted cascade structure is designed to fuse the insights of the deep network and shallow information, optimizing the regression convergence of the defect boundary. It can make full use of the advantages of the deep network and the shallow network, improve the detection accuracy and convergence speed of the defect boundary, enable the detection model to more accurately locate the defect position of the ballast cover plate, and provide more accurate information for subsequent repair work. It can dynamically adjust the detection model to adapt to image changes at different time points and in different environments, improving the real-time performance and accuracy of detection. This makes the entire detection process more efficient and reliable, providing strong support for the anomaly detection of the ballast cover plates of railway tracks.

[0022] The present invention is further configured as follows: in step S5, the generation of candidate regions of anomalies by the region proposal network includes: Extracting implicit semantic features from the original input image according to the convolutional neural network; The region proposal network is learned by a supervised learning method to generate a set of object proposals at each position of the feature map, and the object proposals are used as candidate regions of anomalies; Classifying and performing position regression on the candidate regions generated by the region proposal network, and learning the features of railway images by training a large amount of data.

[0023] In this technical solution, the Region Proposal Network (RPN) is learned through a supervised learning method. At each position on the feature map, a set of object proposals are generated as candidate regions for anomalies, which can efficiently generate candidate regions and avoid the computational complexity of exhaustively searching all candidate regions for anomalies in traditional methods, greatly improving the detection efficiency. The object proposals generated by the RPN are based on the implicit semantic features extracted from the original input image by a convolutional neural network, which can capture the key information in the image and make the generated object proposals more accurately locate the candidate regions for anomalies. This helps the subsequent classification and location regression steps to more accurately judge anomalies. By classifying and performing location regression on the candidate regions generated by the RPN and training a large amount of data to learn the features of railway images, the classification and location regression can be made more accurate, thereby improving the detection accuracy. At the same time, since the candidate regions have been preliminarily screened, subsequent processing only needs to focus on these regions, further improving the detection precision. The RPN generates candidate regions through a supervised learning method, enabling this method to adapt to the detection of anomalies in railway track bed covers under different scenarios. Whether it is different lighting conditions, shooting angles, or types of track bed covers, this method can generate accurate candidate regions, providing a reliable basis for subsequent detection.

[0024] The present invention is further configured such that: the optical axis of the camera forms a certain angle with the track bed plane to ensure that the fields of view of the two cameras completely cover the track bed cover area. The calculation formula for the angle between the optical axis of the camera and the track bed plane is: ; where θ is the angle between the optical axis of the camera and the track bed plane, W is the width of the track bed, and H is the height of the camera relative to the track bed plane when the train is running.

[0025] Advantages of the present invention: (1) Obtain the depth value of the track bed cover plate, identify abnormal warping of the cover plate, perform real-time compensation processing on the image, and improve the accuracy and efficiency of detection; (2) Synchronously collect images of the track area from at least two cameras at different angles, simulate the principle of binocular vision, accurately calculate the depth value of the track bed cover plate, construct a fine three-dimensional model, provide rich three-dimensional spatial data for subsequent anomaly detection, help to more comprehensively understand the state of the track bed cover plate. During the high-speed train operation, perform real-time compensation processing on the image, effectively eliminate motion blur and enhance contrast, ensure that clear and high-quality images can be obtained even in a high-speed motion environment. Set a periodic action sequence including different light intensities, focal length adjustments, and shooting frame rate changes, and set the duration of each action according to the train running speed and the camera shooting frequency, so that the detection method can adapt to different environmental conditions, improve the robustness and applicability of the detection. Extract the feature points of the image under each action, perform feature matching and depth value fusion, and can accurately determine whether the anomaly is due to the track bed cover plate itself, camera shooting parameters, or train running speed changes. Considering multiple factors comprehensively, improve the accuracy and comprehensiveness of anomaly judgment. Based on the judgment of the anomaly result, construct a residual image law model, and use a deep learning network to perform real-time correction and analysis on subsequent images. Generate candidate regions of anomalies through a region proposal network, and perform classification and position regression, which can achieve accurate positioning and classification of anomalies such as missing and warping of the track bed cover plate, not only improve the accuracy of detection, but also significantly improve the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flow chart of the method for detecting abnormal track bed cover plates of the railway track of the present invention; Figure 2 It is a schematic structure diagram of the track bed cover plate in the present invention Figure 1 ; Figure 3 It is a schematic structure diagram of the track bed cover plate in the present invention Figure 2 ; Figure 4 It is a schematic structure diagram of the track bed cover plate in the present invention Figure 3 . DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described here are only the best embodiments of the present invention, only used to explain the present invention, and do not limit the protection scope of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0028] As Figures 1 to 4As shown in the figure, as a first embodiment of the present invention, a method for detecting abnormality of a roadbed cover of a railway track is characterized by comprising the following steps: S1, using at least two cameras to synchronously collect images of the track area from different angles, calculating the depth value of the track bed cover according to the parallax of the same target track area in the imaging of different cameras, and constructing a three-dimensional model of the track bed cover; S2, when the train is running at high speed, the image is compensated in real time; S3, setting a periodic action sequence including different illumination intensities, focus adjustment and shooting frame rate changes, setting the duration of each action according to the train speed and camera shooting frequency, and collecting multiple sets of track images during the duration of each action; S4, extracting feature points of the image under each action, performing feature matching and depth value fusion, and determining whether the abnormality is caused by the roadbed cover body, camera shooting parameters or train speed changes; S5, based on the abnormal judgment results, builds an afterimage law model, performs real-time correction on subsequent images, generates abnormal candidate regions through the region proposal network, classifies and regresses the candidate regions, determines whether the roadbed cover is missing or lifted, and locates it.

[0029] In this embodiment, at least two cameras are used to synchronously collect images of the track area from different angles, and a stereo matching algorithm is used to simulate the binocular vision principle, so that the depth value of the roadbed cover can be accurately calculated, and a fine three-dimensional model can be constructed, which provides rich three-dimensional spatial data for subsequent abnormality detection and helps to understand the state of the roadbed cover more comprehensively. During the high-speed running of the train, the image is compensated in real time, which effectively eliminates motion blur and enhances contrast, ensuring that clear and high-quality images can be obtained even in a high-speed motion environment. A periodic action sequence including different light intensities, focal length adjustments and shooting frame rate changes is set, and the duration of each action is set according to the train speed and camera shooting frequency, so that the detection method can adapt to different environmental conditions and improve the robustness and applicability of the detection. The feature points of the image under each action are extracted, and feature matching and depth value fusion are performed, so that it can accurately determine whether the abnormality is caused by the roadbed cover body, camera shooting parameters or train speed changes. A variety of factors are comprehensively considered to improve the accuracy and comprehensiveness of abnormality judgment. Based on the judgment of abnormal results, a residual image law model is constructed, and the deep learning network is used to perform real-time correction and analysis on subsequent images. The abnormal candidate areas are generated through the region proposal network, and classification and position regression are performed. It can achieve accurate positioning and classification of abnormalities such as missing and lifted roadbed covers, which not only improves the detection accuracy, but also significantly improves the detection efficiency.

[0030] In one embodiment of the present invention, the method further comprises camera setting optimization, wherein the camera setting optimization is: Set a camera to take pictures at three different positions. Select the positions according to the layout of the track bed cover plate and the candidate areas of anomalies. Control the switching of the camera at different positions through a mechanical device. After each switching, use a calibration device to calibrate the internal and external parameters of the camera by taking pictures of a calibration plate with a known size. Compare the differences in the images at different positions, use an image stitching algorithm to stitch the images, and analyze the differences in the images at different positions in the stitched image to assist in judging the cause of the anomaly. Or, set three cameras at the same position to take pictures from different angles or with different parameters; when taking pictures from different angles, the angles are selected to ensure obtaining different side information of the cover plate in the track area. When taking pictures with different parameters, the parameters include exposure time, focal length, and aperture. Use automatic control and synchronous shooting technology to ensure that the time bases of the images are the same. When comparing the images, use image registration and fusion technology to decompose the images into sub-bands of different scales. According to the characteristics of each scale sub-band of the images taken by different cameras, use a weighted average method to fuse them, establish a correspondence relationship model between image differences and anomaly factors, and determine the cause of the anomaly based on the difference points.

[0031] In this technical solution, a mechanical device is used to control a single camera to switch among three different positions, forming a multi-position camera switching strategy to ensure that the camera can comprehensively cover each area of the track bed cover plate, especially the areas prone to anomalies. After each switching, a calibration device is used to take pictures of a calibration plate with a known size to accurately calibrate the internal and external parameters of the camera, thus ensuring that the images taken at different positions have a consistent geometric relationship, laying a solid foundation for subsequent image stitching and difference analysis. An image stitching algorithm is used to seamlessly stitch the images taken at different positions into a complete image, facilitating the inspectors to comprehensively observe the state of the track bed cover plate. At the same time, by analyzing the differences in the images at different positions in the stitched image, the cause of the anomaly can be more accurately assisted in judgment, improving the accuracy and efficiency of detection.

[0032] Set three cameras at the same position to take pictures from different angles or with different parameters (such as exposure time, focal length, aperture, etc.), forming a multi-camera array strategy to obtain more comprehensive and diverse information about the track bed cover plate. The multi-angle and multi-parameter shooting methods help to capture the subtle changes of the track bed cover plate, providing more clues for anomaly detection. Image registration and fusion technology are used to accurately align and fuse the images taken by different cameras. The fused image not only retains the useful information taken by each camera but also fuses the characteristics of different scale sub-bands through a weighted average method, further enriching the image features and improving the accuracy of anomaly detection. A correspondence relationship model between image differences and anomaly factors is established, and the cause of the anomaly is determined based on the difference points in the fused image, which can comprehensively consider various factors such as light changes and shooting angle differences, so as to more accurately judge whether the anomaly is caused by the track bed cover plate itself or other external factors.

[0033] It is understandable that image registration is a process of matching and superimposing two or more images obtained at different times, by different sensors (imaging devices), or under different conditions (weather, illumination, camera position and angle, etc.). The specific steps include feature detection, matching, transformation model calculation, and image resampling. Feature detection can be manual or automatic detection of significant and unique objects (such as closed boundary regions, edges, contours, intersection lines, corner points, etc.). Feature matching is to establish the correlation between the features of the scene image and the reference image. The transformation model calculation is used to calculate the type and parameters of the mapping function for aligning the sensed image and the reference image. Image resampling is to transform the sensed image using the mapping function and calculate the image values at non-integer coordinates using appropriate interpolation techniques.

[0034] It is understandable that image fusion is to integrate the image data of the same target collected from multiple source channels, through image processing and computer technology, etc., to extract the favorable information in each channel to the greatest extent, and finally synthesize high-quality images. The purpose of image fusion is to improve the image quality and interpretation accuracy, and improve the reliability of computer interpretation and the spatial and spectral resolutions of the original images. The fusion methods include multi-sensor image fusion (such as visible light image and infrared image fusion) and single-sensor multi-focus image fusion.

[0035] In one embodiment of the present invention, in step S1, the calculating the depth value of the track bed cover according to the parallax in the imaging of different cameras in the same target orbit region includes: dividing the image obtained by one of the cameras into multiple small image blocks, searching for the most similar region in the image of the other camera, obtaining the pixel point parallax by calculating the displacement of the image blocks, calculating the depth value according to the principle of triangulation, and constructing a three-dimensional model of the track bed cover.

[0036] In this technical solution, the image obtained by one of the cameras is divided into multiple small image blocks, and the most similar region is searched in the image of the other camera. By calculating the displacement of the image blocks, the parallax of the pixel points is obtained, and then the depth value of the track bed cover plate is accurately calculated according to the principle of triangulation. This method based on the stereo matching algorithm can make full use of the principle of binocular vision to improve the accuracy and reliability of depth value calculation. Using the calculated depth value, a three-dimensional model of the track bed cover plate can be constructed, which can intuitively display the shape, position and spatial relationship of the track bed cover plate, providing rich three-dimensional space data for subsequent anomaly detection. Through the three-dimensional model, the inspectors can more comprehensively understand the state of the track bed cover plate, which helps to discover potential anomalies or defects. Based on the accurately calculated depth value and the constructed three-dimensional model, the fine detection of the track bed cover plate can be realized. By comparing the three-dimensional model with the model in the normal state, it is easier to find the anomalies or deformations of the track bed cover plate, which not only improves the accuracy of anomaly detection, but also helps to discover potential safety hazards in advance. The method of simulating the principle of binocular vision by the stereo matching algorithm has strong robustness and can work stably under different lighting conditions, shooting angles and distances. This robustness enables the method to adapt to various complex railway track environments and improves the stability and reliability of the system.

[0037] It can be understood that in stereo matching, the most similar region means that most of the points visible in one view are also visible in the other view, and the matching image regions are similar in appearance. To find the most similar region, a dense correlation method is usually used, that is, for each pixel, the most similar matching point is found in the other image, which can be achieved by calculating the similarity of the small window (patch) around the pixel, such as using the sum of squared differences or the cross-correlation method.

[0038] It can be understood that triangulation is a key technology for obtaining three-dimensional information in computer vision. It uses epipolar geometry and homography matrix to solve the camera motion and the depth of feature points. Specifically, for the matching feature points in two images, their coordinates in the three-dimensional space can be calculated according to the principle of triangulation. The principle of triangulation involves epipolar geometry constraints and homography matrix, and usually the depth of feature points needs to be approximated by the least squares method because in actual operation, due to the influence of noise, the two lines of sight (the lines connecting the two camera centers to the feature points) often cannot intersect precisely.

[0039] It can be understood that the method of establishing the 3D model of the roadbed cover is similar to that of the rail, and can be constructed by reconstructing the 3D entity technology from the 2D contour line. The key is to find the position of the roadbed cover, which can be obtained based on the position of the line centerline and the track gauge. By obtaining the 2D contour line of the roadbed cover and using the 3D reconstruction technology in computer graphics, a 3D model of the roadbed cover can be constructed, so as to more intuitively display and analyze the structure and status of the roadbed cover.

[0040] In one embodiment of the present invention, in step S2, performing real-time compensation processing on the image during high-speed running of the train includes: The algorithm based on motion blur model is used to calculate the degree of image motion blur according to the train speed and camera shooting parameters, and the blurred image is restored through inverse filtering method; Feature points are extracted from continuous frame images through feature point detection algorithm, and the motion vectors between adjacent frames of feature points are calculated according to the optical flow method to obtain the overall motion information of the image. The blurred image is reversely compensated based on the overall motion information of the image, and the bilinear interpolation algorithm is used for resampling. The image is processed by region according to the adaptive histogram equalization algorithm.

[0041] In this technical solution, the algorithm based on the motion blur model, combined with the train speed and camera shooting parameters, can accurately calculate the degree of motion blur of the image. Through the inverse filtering method, the blurred image is restored, which effectively eliminates the image blur caused by the high-speed movement of the train, making the image edge clearer and the details richer. The feature point detection algorithm is used to extract feature points in continuous frame images, and the motion vector between adjacent frames of the feature points is calculated by the optical flow method to obtain the overall motion information of the image. Based on this information, the blurred image is reversely compensated to correct the image distortion caused by motion blur. Subsequently, the bilinear interpolation algorithm is used to resample the image, further smooth the image, eliminate the jagged or distortion phenomenon generated during the compensation process, and significantly improve the image quality. The adaptive histogram equalization algorithm is used to process the image in different regions. This algorithm can automatically adjust the contrast according to the brightness distribution of different regions of the image, making the dark details in the image clearer and the bright details richer. At the same time, the regional processing method can avoid the noise amplification and detail loss problems caused by global histogram equalization, so that the image contrast can be improved more naturally and evenly. By enhancing the contrast of the image, the texture, edge and other features of the roadbed cover are more prominent, which helps the subsequent anomaly detection algorithm to more accurately identify abnormal phenomena such as missing and warping of the roadbed cover. At the same time, the improvement of contrast also enables the inspection personnel to observe the image more clearly, improving the efficiency and accuracy of the inspection.

[0042] It is understandable that algorithms based on the motion blur model are used to simulate and process image blurring caused by the rapid movement of objects. Motion blur algorithms include linear motion blur, rotational motion blur, and scaling motion blur, etc. These algorithms are usually based on the characteristic of visual inertia, that is, when the light acting on the human eye suddenly disappears, the brightness sensation does not disappear immediately, but gradually disappears approximately according to an exponential law. By simulating this motion blur effect, image deblurring can be performed or images with motion blur effects can be generated.

[0043] It is understandable that there are various methods for calculating the degree of image motion blur, including the Fourier transform method, wavelet transform method, Laplacian operator method, and gradient method, etc. The Fourier transform method realizes high-pass filtering by masking the central region of the spectrogram, retains high-frequency information such as image edges, and then calculates the mean value of the spectrogram to evaluate the degree of blur. The wavelet transform method uses wavelet transform to perform multi-scale decomposition on the image, extracts high-frequency detail information, and judges the degree of blur by analyzing the energy distribution of these detail information. The Laplacian operator method and the gradient method respectively reflect the edge information by calculating the second-order derivative map and gradient of the image, and then evaluate the degree of blur.

[0044] It is understandable that feature point detection algorithms are used to detect points with significant features in an image, such as corner points, edge points, etc. Common feature point detection algorithms include FAST, SURF, and SIFT, etc. The FAST algorithm detects corner points by comparing the magnitudes of pixel brightness and has the characteristic of fast speed. The SURF and SIFT algorithms extract feature points by constructing a scale pyramid and calculating the main direction of key points, and have scale invariance and rotation invariance, which are suitable for detecting image feature points with different scales and rotation angles.

[0045] It is understandable that the optical flow method is a technique in computer vision used to calculate the motion of each pixel point in an image between consecutive frames. The Lucas-Kanade optical flow algorithm is a commonly used optical flow algorithm. It calculates the motion vector of pixels by matching the image within a local window and assuming that the pixels within the window have the same motion. This algorithm calculates the motion vector of pixels within the window by minimizing an error function, usually involving solving a system of linear equations. The optical flow method has wide applications in fields such as motion calculation, object tracking, and video compression.

[0046] It is understandable that the bilinear interpolation algorithm is an image interpolation algorithm used to calculate the values of new image pixels during image scaling, rotation, etc. This algorithm uses the pixel values of the four nearest neighbor points in the original image and calculates the values of new image pixels through the bilinear interpolation formula. The bilinear interpolation algorithm is widely used in image processing and computer vision and can better maintain the smoothness and continuity of the image.

[0047] It is understandable that the adaptive histogram equalization algorithm is a technique for image enhancement. By calculating the local histogram of an image and redistributing the brightness, it changes the image contrast. Different from the traditional histogram equalization method, adaptive histogram equalization equalizes according to the local regions of the image, thus avoiding the problems of over-enhancement or distortion caused by global equalization. This algorithm can significantly enhance the contrast of local regions, retain image details, and is applicable to images with uneven contrast. In the abnormal detection method of the roadbed cover plate of railway tracks, the adaptive histogram equalization algorithm can be used to enhance the collected roadbed cover plate images, improve the contrast and clarity of the images, and thus more accurately detect the abnormalities of the roadbed cover plate.

[0048] It is understandable that restoring a blurred image through the inverse filtering method includes the following steps: Perform a Fourier transform on the blurred image to convert it from the spatial domain to the frequency domain, obtaining the blur function; Calculate the Fourier transform of the blur function; In the frequency domain, by dividing the blur function by the Fourier transform of the blur function, an estimate of the Fourier transform of the original image is obtained; Perform an inverse Fourier transform on the estimate of the Fourier transform of the original image to convert it back to the spatial domain, obtaining the restored image.

[0049] In step S3, the extraction of feature points of the image under each action, and the feature matching and depth value fusion include: using the image feature extraction and matching algorithm to process the images taken under different actions, extracting feature points and calculating the descriptors between feature points, and matching the feature points by comparing the similarity of descriptors; combining the depth values to analyze the differences in feature points between images of different actions.

[0050] In this technical solution, by using the image feature extraction and matching algorithm, the significant feature points in the captured images under different actions can be accurately extracted, and the descriptors between these feature points can be calculated. By comparing the similarity of the descriptors, the feature points in different images can be accurately matched, providing a reliable basis for subsequent depth value fusion and anomaly judgment. Combining the depth values, the differences in feature points between images of different actions are analyzed. The depth values provide the three-dimensional coordinates of the feature points, which helps to understand the positions and relationships of the feature points in space. By analyzing the differences in feature points, it can be determined whether these differences are caused by changes in the ballast cover body itself, such as anomalies like missing or warping, or are caused by factors such as camera shooting parameters and changes in train driving speed. This depth value fusion analysis method improves the accuracy and comprehensiveness of anomaly judgment. Based on the results of feature matching and depth value fusion, it can be accurately determined whether there is an anomaly in the ballast cover and the location of the anomaly can be located. This not only improves the accuracy of anomaly detection but also helps to take timely measures for repair, thus ensuring the safe operation of railway tracks. The feature extraction and matching algorithm has high computational efficiency and can quickly process a large amount of image data. At the same time, by analyzing in combination with depth values, the detection method has stronger robustness to factors such as different lighting conditions, shooting angles, and distances, and can work stably in various complex railway track environments.

[0051] Preferably, in step S4, the determination that the anomaly is due to changes in the ballast cover body, camera shooting parameters, or train driving speed includes: By comparing the changes in the matching values and depth values of feature points under different action sequences, if the feature point matching fails or there are local depth mutations or surface unevenness in the depth values, it is determined that there is an anomaly in the ballast cover body; Analyze the influence of different exposure times, focal lengths, and apertures on image features. If the feature point matching is successful but there is an overall depth shift or image blurring after depth value fusion, it is determined that there is an anomaly in the camera shooting parameters; According to the train driving speed and the camera shooting frequency, set the duration of the action sequence. If there is motion blurring or image misalignment in the results of feature point matching and depth value fusion, it is determined that the anomaly is caused by changes in the train driving speed.

[0052] In this technical solution, when the feature point matching fails, or there are local depth mutations or uneven surfaces in the depth values, it is determined that the ballast cover body is abnormal, and physical changes of the ballast cover, such as missing or warping, can be directly detected, providing accurate information for taking timely measures for repair. When the feature point matching is successful but there is an overall depth offset or blurred image after the depth values are fused, it is determined that the camera shooting parameters are abnormal, and image quality problems caused by improper camera parameter settings (such as exposure time, focal length, aperture, etc.) can be identified, which helps to adjust the camera parameters to obtain higher-quality images. According to the train running speed and the camera shooting frequency, the duration of the action sequence is set. When there is motion blur or image misalignment in the feature point matching and depth value fusion results, it is determined that the abnormality is caused by the change in the train running speed, and the influence of the train speed change on the image acquisition and detection process can be detected, which helps to optimize the detection algorithm to adapt to different train running speeds. By comprehensively considering multiple factors to judge the source of the abnormality, false judgments and missed judgments caused by single-factor judgment are avoided, not only improving the accuracy of abnormality judgment, but also making the detection method more comprehensive and reliable.

[0053] In step S5, the constructing the afterimage rule model based on the abnormal judgment result includes: Adopt the BakeMesh method to create an afterimage effect by copying the geometric model of the ballast cover and adjusting its transparency and position at different time points; Use screen post-processing technology to post-process the collected ballast cover images to generate an afterimage effect, and analyze the motion trajectory and state changes of the ballast cover by comparing the afterimage images at different time points; Through the vertex offset method, adjust the vertex positions of the ballast cover geometric model to simulate its state changes at different time points and generate a continuous afterimage effect.

[0054] In this technical solution, the BakeMesh method is adopted. By copying the geometric model of the track bed cover plate and adjusting its transparency and position at different time points to create an afterimage effect, it can intuitively display the state changes of the track bed cover plate at different time points, providing visual support for analysts. Analysts can more clearly observe the movement trajectory and state change trend of the track bed cover plate. Using screen post-processing technology to post-process the collected images of the track bed cover plate to generate an afterimage effect can further enhance the fidelity of the afterimage effect, making the afterimage image closer to the real situation and providing more accurate visual information for analysts. By using the vertex offset method to adjust the vertex positions of the geometric model of the track bed cover plate and simulate its state changes at different time points to generate a continuous afterimage effect, it can more finely control the generation process of the afterimage effect, making the afterimage effect more continuous and natural and providing a smoother viewing experience for analysts. By comparing the afterimage images at different time points, the movement trajectory and state changes of the track bed cover plate can be analyzed, which helps to understand the movement law and state change trend of the track bed cover plate at different time points and provides important data support for subsequent anomaly detection and state assessment. At the same time, by analyzing the afterimage law, the future state changes of the track bed cover plate can also be predicted, providing a basis for taking preventive measures for repair in advance.

[0055] Preferably, in step S5, the real-time correction of subsequent images includes: Performing denoising and contrast enhancement preprocessing on the collected images of the track bed cover plate according to a convolutional neural network; Integrating a multi-head self-attention module into the deep learning network to capture complex features in the images of the track bed cover plate; Designing a cross-layer weighted cascade structure to fuse deep network insights and shallow information and optimize the regression convergence of defect boundaries.

[0056] The preprocessing of denoising and enhancing the contrast of the collected ballast cover plate images through a convolutional neural network effectively eliminates the noise in the images, improves the clarity and contrast of the images, makes the subsequent feature extraction and defect detection more accurate, and provides higher-quality image data for the entire detection process. Integrating the multi-head self-attention module into the deep learning network enables the network to capture the complex features in the ballast cover plate images. The self-attention mechanism can focus on the key information in the images, while the multi-head design can capture features from different aspects, improving the diversity and robustness of feature extraction, helping to more comprehensively understand the state of the ballast cover plate, and improving the accuracy of anomaly detection. Designing a cross-layer weighted cascade structure to fuse the insights of the deep network and shallow information optimizes the regression convergence of the defect boundary, can make full use of the advantages of the deep network and the shallow network, improves the detection accuracy and convergence speed of the defect boundary, enables the detection model to more accurately locate the defect position of the ballast cover plate, and provides more accurate information for the subsequent repair work. It can dynamically adjust the detection model to adapt to the image changes at different time points and in different environments, improving the real-time performance and accuracy of the detection. This makes the entire detection process more efficient and reliable, providing strong support for the anomaly detection of the ballast cover plates of railway tracks.

[0057] The optical axis of the camera forms a certain angle with the ballast plane to ensure that the fields of view of the two cameras completely cover the ballast cover plate area. The calculation formula for the angle between the optical axis of the camera and the ballast plane is: ; where θ is the angle between the optical axis of the camera and the ballast plane, W is the width of the ballast, and H is the height of the camera relative to the ballast plane when the train is running.

[0058] In step S5, the generation of candidate regions for anomalies by the region proposal network includes: Extracting implicit semantic features from the original input image according to the convolutional neural network; The region proposal network learns through a supervised learning method to generate a set of object proposals at each position of the feature map, and the object proposals serve as candidate regions for anomalies; Classifying and performing position regression on the candidate regions generated by the region proposal network, and learning the features of railway images by training a large amount of data.

[0059] The Region Proposal Network is learned through a supervised learning method. At each position of the feature map, it generates a set of object proposals as candidate regions for anomalies. It can efficiently generate candidate regions, avoiding the computational complexity of exhaustively enumerating all candidate regions for anomalies in traditional methods, and greatly improving the detection efficiency. The object proposals generated by the Region Proposal Network are based on the implicit semantic features extracted from the original input image by a convolutional neural network, which can capture the key information in the image, making the generated object proposals more accurately locate the candidate regions for anomalies. This helps the subsequent classification and location regression steps to more accurately judge anomalies. Classifying and performing location regression on the candidate regions generated by the Region Proposal Network, and learning the features of railway images by training a large amount of data, can make the classification and location regression more accurate, thereby improving the detection accuracy. At the same time, since the candidate regions have been preliminarily screened, subsequent processing only needs to focus on these regions, further improving the detection precision. The Region Proposal Network generates candidate regions through a supervised learning method, enabling this method to adapt to the detection of anomalies in railway track bed covers under different scenarios. Whether it is different lighting conditions, shooting angles, or types of track bed covers, this method can generate accurate candidate regions, providing a reliable basis for subsequent detection.

[0060] The above embodiments are specific descriptions of the present invention, which are only used to further illustrate the present invention and should not be construed as limiting the protection scope of the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the content of the above invention fall within the protection scope of the present invention.

Claims

1. A method for detecting abnormality of a railway track ballast cover, characterized in that: The following steps are involved: S1, using at least two cameras to synchronously collect images of the track area from different angles, calculating the depth value of the track bed cover according to the parallax of the same target track area in the imaging of different cameras, and constructing a three-dimensional model of the track bed cover; S2, when the train is running at high speed, the image is compensated in real time; S3, setting a periodic action sequence including different illumination intensities, focus adjustment and shooting frame rate changes, setting the duration of each action according to the train speed and camera shooting frequency, and collecting multiple sets of track images during the duration of each action; S4, extracting feature points of the image under each action, performing feature matching and depth value fusion, and determining whether the abnormality is caused by the roadbed cover body, camera shooting parameters or train speed changes; S5, based on the abnormal judgment results, builds an afterimage law model, performs real-time correction on subsequent images, generates abnormal candidate regions through the region proposal network, classifies and regresses the candidate regions, determines whether the roadbed cover is missing or lifted, and locates it.

2. A method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: The method further includes camera setting optimization, wherein the camera setting optimization is: Set a camera to shoot at three different positions, select the position according to the layout of the roadbed cover and the candidate area of ​​the anomaly, control the camera to switch between different positions through a mechanical device, use the calibration device after each switch, shoot the image of the calibration plate of known size, calibrate the camera's internal and external parameters, compare the image differences at different positions, use the image stitching algorithm to stitch the images, and analyze the image differences at different positions in the stitched image to assist in determining the cause of the anomaly; Alternatively, three cameras are set at the same position to capture images from different angles or with different parameters. When capturing images at different angles, the angle selection ensures that information on different sides of the track area cover is obtained. When capturing images with different parameters, the parameters include exposure time, focal length, and aperture. Automatic control and synchronous shooting techniques are used to ensure that the image time reference is the same. When comparing images, image registration and fusion techniques are used to decompose the image into sub-bands of different scales. According to the characteristics of each sub-band of each scale of images from different cameras, the weighted average method is used for fusion. A model of the correspondence between image differences and abnormal factors is established, and the cause of the abnormality is determined based on the difference points.

3. The method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: In step S1, the depth value of the roadbed cover is calculated according to the parallax of the same target track area in different camera imaging, and the three-dimensional model of the roadbed cover is constructed, including: dividing the image obtained by one of the cameras into multiple small image blocks, searching for the most similar area in the other camera image, obtaining the pixel point parallax by calculating the image block displacement, calculating the depth value according to the triangulation principle, and constructing the three-dimensional model of the roadbed cover.

4. The method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: In step S2, the real-time compensation processing of the image during the high-speed running of the train includes: The algorithm based on motion blur model is used to calculate the degree of image motion blur according to the train speed and camera shooting parameters, and the blurred image is restored through inverse filtering method; Feature points are extracted from continuous frame images through feature point detection algorithm, and the motion vectors between adjacent frames of feature points are calculated according to the optical flow method to obtain the overall motion information of the image. The blurred image is reversely compensated based on the overall motion information of the image, and the bilinear interpolation algorithm is used for resampling. The image is processed by region according to the adaptive histogram equalization algorithm.

5. The method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: In step S3, the feature points of the image under each action are extracted, and the feature matching and depth value fusion are performed, including: using the image feature extraction and matching algorithm to process the images taken under different actions, extracting feature points and calculating the descriptors between the feature points, and matching the feature points by comparing the similarity of the descriptors; combining the depth value to analyze the differences in feature points between images of different actions.

6. The method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: In step S4, the determination that the abnormality is caused by the change in the roadbed cover body, camera shooting parameters or train speed includes: By comparing the changes in the matching values ​​and depth values ​​of feature points under different action sequences, if the feature point matching fails or the depth value has a local depth mutation or the surface is uneven, it is determined that the track bed cover body is abnormal; Analyze the impact of different exposure times, focal lengths, and apertures on image features. If feature point matching is successful but overall depth offset and image blur occur after depth value fusion, it is determined that the camera shooting parameters are abnormal. The duration of the action sequence is set according to the train speed and the camera shooting frequency. If motion blur and image misalignment appear in the feature point matching and depth value fusion results, it is determined to be an abnormality caused by the change in train speed.

7. The method for detecting abnormality of a railway track ballast cover according to claim 1, characterized in that: In step S5, the construction of the afterimage regularity model based on the abnormal result judgment includes: The BakeMesh method is used to create the afterimage effect by duplicating the geometric model of the roadbed cover and adjusting its transparency and position at different time points; The collected images of the roadbed cover are post-processed using screen post-processing technology to generate afterimage effects. The movement trajectory and state changes of the roadbed cover are analyzed by comparing the afterimage images at different time points. The vertex positions of the geometric model of the roadbed cover are adjusted through the vertex offset method to simulate its state changes at different time points and generate a continuous afterimage effect.

8. The method for detecting abnormality of a railway track ballast cover according to claim 7, characterized in that: In step S5, the real-time correction of the subsequent image includes: The collected roadbed cover images are preprocessed by denoising and enhancing contrast according to the convolutional neural network; Integrate a multi-head self-attention module into a deep learning network to capture complex features in the roadbed cover image; A cross-layer weighted cascade structure is designed to integrate deep network insights with shallow information and optimize the regression convergence of defect boundaries.

9. A method for detecting abnormality of a railway track ballast cover according to claim 8, characterized in that: In step S5, the generation of abnormal candidate regions by the region proposal network includes: Extract implicit semantic features from the original input image based on the convolutional neural network; The region proposal network is learned through a supervised learning method to generate a set of target proposals at each position of the feature map, and the target proposals are used as candidate regions of abnormalities; The candidate regions generated by the region proposal network are classified and regressed, and the characteristics of railway images are learned by training a large amount of data.

10. A method for detecting abnormality of a railway track ballast cover according to any one of claims 1 to 9, characterized in that: The camera optical axis forms a certain angle with the roadbed plane to ensure that the field of view of the two cameras completely covers the roadbed cover area. The angle between the camera optical axis and the roadbed plane is calculated as follows: ； Among them, θ is the angle between the camera optical axis and the track bed plane, W is the width of the track bed, and H is the height of the camera relative to the track bed plane when the train is running.

Images