Railway track area chipping detection method and system

Abstract

The invention discloses a railway track area chipping detection method which comprises the following steps: a track area image acquisition step: acquiring a track area image of a travelling crane as a sleeper image through a visual camera; a chipping image detection step: chipping features in the sleeper image are extracted based on a preset image edge detection algorithm model, and the image edge detection algorithm model is configured to fuse Canny edge detection and multi-scale gradient feature extraction; a falling block verification step: when the falling block feature is detected, starting a falling block verification device to jet airflow to the area where the falling block feature is located so as to drive the falling block to move; an image continuous acquisition step: continuously acquiring a track area image as a verification image after the falling block verification device is started; and a block falling position comparison and output step: registering the sleeper image and the verification image, dividing a coordinate system by taking the two rails as references, comparing the relative position offset of the block falling characteristics in the coordinate system of the two rails, and if the offset exceeds a preset threshold value, judging that the block falling is real and outputting a marking result.

Description

Technical Field

[0001] The present invention relates to the field of railway track detection, and particularly to a method and system for detecting rail surface spalling in a railway track area. Background Art

[0002] With the rapid development of China's rail transit industry, the mileage of rail laying has been increasing year by year. In rail transit, the repeated rolling of train wheels can cause various hazards on the rail surface, such as wear, joint collapse, spalling, and track cracking. Among them, the spalling problem is particularly serious. If the spalling position cannot be detected and repaired in time, the formed notch will lead to increased wheel impact and even derailment accidents. In addition, delayed treatment may also result in the need to replace an entire section of the rail, and the cost is much higher than local repair.

[0003] Currently, the main methods for detecting spalling include manual inspection and using track inspection vehicles. Track inspection vehicles are usually equipped with devices such as high-definition high-speed cameras, laser scanners, or ultrasonic flaw detectors to detect spalling or the position of spalling in the track area. Among them, the high-definition high-speed camera takes continuous photos of the track area and uses image processing algorithms to detect whether there is spalling. This method mainly relies on identifying objects in the image that have a color difference from the sleeper and a regular shape, and marking them as potential spalling.

[0004] Existing spalling detection technologies have some obvious limitations: First, in areas with complex terrain, such as subway systems in mountainous or coastal areas, the track may switch between underground and ground. In the underground environment, due to fewer interference factors, the accuracy of spalling recognition by image algorithms is relatively high. However, in the ground situation, the environment is more complex, and rain, bird droppings, etc. may fall into the track area, seriously interfering with the recognition of image algorithms, making it difficult to identify spalling on open-air tracks. Second, existing image recognition algorithms mainly rely on color difference and shape regularity to judge spalling. This method is easily affected by environmental factors such as light changes, shadows, and stains, resulting in false positives or false negatives. Simply relying on image recognition cannot determine whether the detected anomaly is a real spalling or other foreign objects, which may lead to unnecessary manual inspections, increasing maintenance costs and workload. In addition, existing technologies lack the ability to quantitatively analyze spalling and cannot accurately estimate the size, position, and potential hazard degree of spalling, which is crucial for formulating effective maintenance strategies.

[0005] Existing detection systems lack a real-time verification mechanism. Once a suspected spalling is found, it usually requires arranging a manual inspection for confirmation, which is not only time-consuming and laborious but also may delay the timely treatment of real spalling. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for detecting block loss in the railway track area, which has the advantages of improving the accuracy of block loss detection, reducing false positives and missed detections, realizing real-time verification, and providing quantitative analysis capabilities.

[0007] This application provides a method for detecting block loss in the railway track area. The technical solution is as follows: It includes the following steps: a track area image acquisition step, where a visual camera is used to collect the track area image of the running train as the sleeper image; a block loss image detection step, where the block loss features in the sleeper image are extracted based on a preset image edge detection algorithm model, and the image edge detection algorithm model is configured to fuse Canny edge detection and multi-scale gradient feature extraction; a block loss verification step, when the block loss features are detected, a block loss verification device is started. The block loss verification device includes an adjustable jet group for jetting air flow to the area where the block loss features are located to drive the block to move; an image continuous acquisition step, after the block loss verification device is started, continuously collect the track area image as the verification image; a block loss position comparison and output step, register the sleeper image and the verification image, divide the coordinate system based on the two tracks, compare the relative position offset of the block loss features in the coordinate systems of the two tracks, and if the offset exceeds the preset threshold, it is determined as a real block loss and the marking result is output.

[0008] Further, this application also proposes that both the sleeper image and the verification image are processed by a zoning algorithm, divided into several parallel zones with the two tracks as the boundaries. Each zone is parallel to the track direction and the width is adjustable, and each zone corresponds to an independent position code; when the block loss features are detected in a certain zone, the jet group of the block loss verification device is adjusted directionally to match the spatial coordinates of this zone.

[0009] Further, this application also proposes that the block loss verification device includes at least two groups of jet groups, which are respectively deployed on both sides of the track and symmetrically distributed on both sides of the track center line; each group of jet groups includes a jet nozzle, a swing component for driving the deflection angle of the jet nozzle, and a gas source for providing high-pressure gas to the jet nozzle. The block loss verification device is configured with a gas path control system; when the block loss features are located in a zone on one side of the track center line, the jet group close to this side is started, and the swing component adjusts the jet angle of the jet nozzle according to the zone coordinates, and the jet nozzle drives the air flow to act on the block loss feature area.

[0010] Further, this application also proposes that the zone width of the verification image is strictly equal to the zone width of the sleeper image, and the zone boundaries are aligned by an image registration algorithm; in the block loss position comparison and output step, if the offset of the block loss features in the verification image relative to the sleeper image crosses at least one zone, it is determined as an effective movement, and a block loss movement trajectory report is generated in combination with the air pressure parameters of the jet group and the zone offset distance.

[0011] Furthermore, the present application also proposes a step of verifying the image features of the chipped area, specifically: extracting the edge contours of the chipped area features in the sleeper image and the verification image respectively, and calculating the contour similarity and the area change rate between the two; if the contour similarity is lower than the preset value and the area change rate does not exceed the threshold, it is determined as a static foreign object rather than a chipped area; if the contour similarity meets the requirements and the area change rate is associated with the position offset, it is determined as a real chipped area.

[0012] Furthermore, the present application also proposes a step of storing and classifying the chipped area data, specifically: establishing a chipped area feature database to store the edge contours, movement trajectories, air pressure parameters and partition offset data of real chipped areas; performing clustering analysis on the chipped area contour features through a convolutional neural network to generate chipped area type labels classified by shape and size, and matching with historical data to optimize the determination threshold.

[0013] Furthermore, the present application also proposes a step of secondary verification of the chipped area, specifically: when the jet verification execution module in the first chipped area verification step is started, if the chipped area feature in the verification image does not move, a secondary verification mode is triggered; in the secondary verification mode, the air circuit control system raises the air pressure of the jet group to the preset high pressure threshold and adjusts the angle of the jet nozzle through the swing assembly for secondary spraying; after the secondary spraying, a secondary verification image is obtained through the image continuous acquisition step. If the chipped area feature still does not move, it is determined as a static foreign object, and this feature is marked as a non-chipped area type; and a foreign object database is established to store the position, air pressure parameters and image features of the static foreign object, and synchronously updated to the railway operation and maintenance management system to eliminate the interference of false detection.

[0014] Furthermore, the present application also proposes that in the chipped area image detection step, it also includes calculating the pixel area, aspect ratio and geometric center coordinates of the chipped area feature in the image, and performing spatial mapping in combination with the actual size of the track to output the estimated physical size of the chipped area and the coordinate of the chipped area dropping position.

[0015] Furthermore, the present application also proposes that the chipped area image detection step is configured with a dynamic air pressure adaptation algorithm, specifically including: a chipped area area quantization sub-step, calculating the pixel area S of the chipped area feature in the image through an edge detection processing module, and converting it into a physical area S1 based on the mapping relationship between the actual size of the track and the image resolution; an air pressure calculation sub-step, setting a basic air pressure threshold P0, and dynamically adjusting the first chipped area verification spraying air pressure value P according to the following formula: , where S2 is a preset reference area threshold, α is an area weight coefficient, and the value range is 0.2 ≤ α ≤ 0.5; the secondary spraying air pressure in the secondary verification mode is 1.2 to 2 times that of p.

[0016] Furthermore, the present application also proposes a detection system for block loss in a railway track area, including: an orbital image acquisition module for obtaining sleeper images and verification images through a vision camera; an edge detection and processing module configured to run an image edge detection algorithm model to extract block loss features; a jet verification execution module including symmetrically arranged jet groups, a swing drive unit, and a pressure regulation unit for directionally driving the block loss to move; an image analysis and comparison module for registering images, dividing partitions, and calculating position offsets; and a data storage and output module for storing block loss feature data and generating a detection report.

[0017] As can be seen from the above, a detection method and system for block loss in a railway track area provided by the present application include steps such as obtaining orbital area images, detecting block loss images, verifying block loss, continuously acquiring images, and comparing and outputting the positions of block loss. By combining image processing technology and physical verification methods, accurate identification and verification of block loss are achieved, effectively solving the problems of false judgment and missed judgment existing in the prior art, improving the accuracy of block loss detection, and being able to perform real-time verification and provide quantitative analysis, having the advantages of improving the accuracy of block loss detection, reducing false judgment and missed judgment, realizing real-time verification, and providing quantitative analysis capabilities. Description of the Drawings

[0018] Figure 1 is an overall flowchart of a detection method for block loss in a railway track area; Figure 2 is a sleeper image of a normal railway track area in the present invention; Figure 3 is a sleeper image containing block loss features in the present invention. Detailed Embodiments

[0019] The present invention will be further described in detail below in conjunction with the drawings and embodiments.

[0020] Embodiment: The present application provides a detection method for block loss in a railway track area, including the following steps: combining Figure 1, Rail region image acquisition step: Collect the rail region image of the train by a vision camera as the sleeper image; Chipped image detection step: Extract the chipped features in the sleeper image based on a preset image edge detection algorithm model, and the image edge detection algorithm model is configured to fuse Canny edge detection and multi-scale gradient feature extraction; Chipped verification step: When chipped features are detected, start the chipped verification device, which includes an adjustable jet group for jetting air flow to the area where the chipped features are located to drive the chips to move; Image continuous acquisition step: After the chipped verification device is started, continuously collect the rail region images as verification images; Chipped position comparison and output step: Register the sleeper image and the verification image, divide the coordinate system based on two rails, compare the relative position offset of the chipped features in the coordinate systems of the two rails, and if the offset exceeds the preset threshold, determine it as a real chip and output the marking result of the real chip.

[0021] In the daily maintenance of railway tracks, the detection and repair of chips are an important link. If chips are not repaired in time, it will cause wheel impacts during train operation and even lead to derailment accidents. The existing detection methods mainly rely on manual inspections and track inspection vehicles. However, in places with complex terrains, especially open-air tracks, the accuracy and efficiency of detection are greatly affected by environmental interference. This application proposes a chip detection method based on a vision camera and an image edge detection algorithm. Through multi-step image processing and verification, it can effectively improve the accuracy and reliability of chip detection.

[0022] In the rail region image acquisition step, the rail region image of the train is collected by a vision camera as the sleeper image. In the chipped image detection step, the chipped features in the sleeper image are extracted based on a preset image edge detection algorithm model, and the image edge detection algorithm model fuses Canny edge detection and multi-scale gradient feature extraction. In the chipped verification step, when chipped features are detected, start the chipped verification device, which includes an adjustable jet group for jetting air flow to the area where the chipped features are located to drive the chips to move. In the image continuous acquisition step, after the chipped verification device is started, continuously collect the rail region images as verification images. In the chipped position comparison and output step, register the sleeper image and the verification image, divide the coordinate system based on two rails, compare the relative position offset of the chipped features in the coordinate systems of the two rails, and if the offset exceeds the preset threshold, determine it as a real chip and output the marking result of the real chip.

[0023] The image of the track area is collected by a vision camera, and the block loss feature is extracted based on the image edge detection algorithm model. The block loss verification device drives the block loss to move by jetting air flow, and continuously collects images to verify the actual movement of the block loss. By comparing the positions of the block loss features in the sleeper image and the verification image, it is judged whether the block loss actually exists. This method can effectively improve the accuracy of block loss detection and reduce the influence of environmental interference.

[0024] Compared with the prior art, the method of the present application can detect the block loss feature in the track area more accurately through multi-step image processing and verification. The existing manual inspection and track inspection vehicle have low detection accuracy in complex environments, while the method of the present application can effectively reduce environmental interference and improve the reliability of detection through image edge detection and jet verification.

[0025] The present application collects the image of the track area by a vision camera, extracts the block loss feature based on the image edge detection algorithm model. When the block loss feature is detected, the block loss verification device is started to jet air flow to the area where the block loss feature is located to drive the block loss to move. Subsequently, the images of the track area are continuously collected as verification images. By comparing the positions of the block loss features in the sleeper image and the verification image, it is judged whether the block loss actually exists. Through this method, the accuracy and reliability of block loss detection can be effectively improved, and the influence of environmental interference can be reduced.

[0026] Furthermore, the present application also proposes that Figure 2 Both the sleeper image and the verification image are processed by a zoning algorithm, divided into several parallel zones with two tracks as the boundaries. Each zone is parallel to the track direction and the width is adjustable, and each zone corresponds to an independent position code; when the block loss feature is detected in a certain zone, the jet group of the block loss verification device is adjusted directionally to match the spatial coordinates of the zone.

[0027] In the context of the present application, the zoning processing of the sleeper image and the verification image is to improve the accuracy and efficiency of block loss detection. By dividing the track image into several parallel zones, the width of each zone can be adjusted as needed, and each zone has an independent position code. This zoning processing method makes the detection and positioning of the block loss feature more accurate.

[0028] In the technical solution of the present application, the jet group of the block loss verification device can be adjusted directionally according to the detected zone where the block loss feature is located. Specifically, when the block loss feature is detected in a certain zone, the jet group will adjust its jetting direction to match the spatial coordinates of the zone, so as to ensure that the jet device can accurately aim at the position where the block loss feature is located for jet verification.

[0029] Through the processing of the above partition algorithm, the block loss verification device can more accurately locate and verify the block loss features, avoiding the possible errors in the jet verification process and improving the reliability and accuracy of block loss detection.

[0030] Compared with the prior art, the technical solution of the present application processes the sleeper image and the verification image through the partition algorithm, making the positioning of the block loss features more accurate. Moreover, the jet groups can be directionally adjusted according to the partition coordinates, thereby improving the accuracy and efficiency of block loss verification. Thus, it can more effectively detect and verify the block loss features in the track area, avoiding the problems of false detection and missed detection caused by inaccurate block loss detection.

[0031] Furthermore, the present application also proposes that the block loss verification device includes at least two groups of jet groups, which are respectively deployed on both sides of the track and symmetrically distributed on both sides of the track center line; each group of jet groups includes a jet nozzle, a swing assembly for driving the deflection angle of the jet nozzle, and a gas source for providing high-pressure gas to the jet nozzle. The block loss verification device is configured with a gas path control system; when the block loss feature is in the partition on one side of the track center line, the jet group close to that side is activated, and the swing assembly adjusts the spraying angle of the jet nozzle according to the partition coordinates, and the jet nozzle drives the airflow to act on the block loss feature area.

[0032] The technical solution of the present application can more accurately verify the block loss features by symmetrically deploying jet groups on both sides of the track and using the swing assembly to adjust the spraying angle of the jet nozzle. Specifically, when a block loss feature is detected in a certain partition, the jet group close to that partition will be activated, and the angle of the jet nozzle will be adjusted through the swing assembly so that the airflow can accurately act on the block loss feature area. In this way, the block is driven to move by the jet to verify whether it is a real block loss.

[0033] Multiple possible implementation manners of the new technical features include: 1. The material selection of the jet nozzle can adopt an alloy material that is resistant to high pressure and corrosion to ensure that it is not easily damaged under the action of high-pressure gas; 2. The swing assembly can be driven by an electric servo motor to achieve precise angle control; 3. The gas source can be configured as a high-pressure gas cylinder or a compressed air system, and the pressure is adjusted through the gas path control system to adapt to different block loss features and partition requirements; 4. The gas path control system can be integrated with a pressure sensor and a control valve to achieve real-time monitoring and adjustment of the air pressure; 5. The system can also be configured with redundant jet groups to improve the reliability and stability of the system.

[0034] The solution of this application improves the accuracy and reliability of block drop verification by symmetrically arranging the jet groups and precisely controlling the angles of the jet nozzles. Compared with the prior art, this application can more effectively verify block drop features, reduce false detections and missed detections, and can adapt to changes in different track environments and block drop features by precisely controlling the air pressure and injection angle, improving the efficiency and safety of track maintenance.

[0035] Further, this application also proposes that the partition width of the verification image is strictly equal to the partition width of the sleeper image, and the partition boundaries are aligned through an image registration algorithm. In the step of comparing and outputting the block drop position, if the offset of the block drop feature in the verification image relative to the sleeper image crosses at least one partition, it is determined as an effective movement, and a block drop movement trajectory report is generated by combining the air pressure parameters of the jet group and the partition offset distance.

[0036] The technical solution of this application aims to improve the accuracy and reliability of block drop detection. By ensuring that the partition widths of the verification image and the sleeper image are strictly equal and aligning the partition boundaries through an image registration algorithm, the accuracy of comparing the block drop position can be ensured. When the position of the block drop feature in the verification image shifts relative to the sleeper image and the offset amount exceeds one partition, the system can accurately determine the effective movement of the block drop. Combining the air pressure parameters of the jet group and the partition offset distance, a block drop movement trajectory report is generated to provide more detailed information about the block drop movement.

[0037] The technical feature that the partition widths of the verification image and the sleeper image are strictly equal can be achieved in the following ways: First, when collecting images, a high-precision vision camera is used to ensure consistent image resolution; Second, when processing images, a unified partitioning algorithm is adopted to divide the images to ensure that the width of each partition is the same; Finally, the partition boundaries are aligned through an image registration algorithm to ensure the consistency of the partition positions between different images.

[0038] The technical feature of generating a block drop movement trajectory report can be achieved in the following ways: After detecting the effective movement of the block drop feature, the movement trajectory of the block drop is calculated by combining the air pressure parameters of the jet group and the partition offset distance. Specifically, the movement distance of the block drop under the action of the jet can be determined according to the air pressure parameters of the jet group, and combined with the partition offset amount in the image, a block drop movement trajectory report is generated. This report can provide detailed information such as the movement path and movement distance of the block drop for further analysis and processing.

[0039] This application improves the accuracy and reliability of spall detection by strictly aligning and verifying the partition widths of the verification image and the tie image, and generating a spall movement trajectory report by combining the air pressure parameters of the jet group and the partition offset distance. Compared with the prior art, this application can more accurately determine the effective movement of spalls and provide detailed spall movement information, thus helping to timely detect and handle spall problems and ensuring the safe operation of rail transit.

[0040] Furthermore, this application also proposes a spall feature image verification step, specifically: Combined with Figure 3 , the edge contours of the spall features in the tie image and the verification image are respectively extracted, the contour similarity and the area change rate between the two are calculated. If the contour similarity is lower than the preset value and the area change rate does not exceed the threshold, it is determined as a static foreign object rather than a spall. If the contour similarity meets the requirements and the area change rate is associated with the position offset, it is determined as a real spall, and the real spall result is output.

[0041] The technical solution of the spall feature image verification step aims to further improve the accuracy of spall detection to avoid misidentifying static foreign objects as spalls. By extracting the edge contours of the spall features in the tie image and the verification image, the shape features of the spalls can be accurately obtained. Calculating the contour similarity and the area change rate can effectively determine whether the spall features have changed, so as to distinguish real spalls from static foreign objects.

[0042] Specifically, classical edge detection algorithms such as Canny edge detection can be used for edge contour extraction. Combining image processing techniques, the edge contours of the spall features are extracted. The contour similarity calculation can use shape matching algorithms such as Hausdorff distance or shape context descriptor to evaluate the similarity degree of the two contours. The area change rate is judged by calculating the area difference of the spall features in the tie image and the verification image.

[0043] For example, if a spall feature is detected in the tie image, after starting the spall verification device, the verification image is obtained, and the edge contour of the spall feature in the verification image is extracted. The contour similarity and the area change rate of the spall feature in the tie image and the verification image are calculated. If the contour similarity is lower than the preset value (such as 0.8), and the area change rate does not exceed the threshold (such as 10%), then this feature is determined as a static foreign object. If the contour similarity meets the requirements (such as greater than or equal to 0.8), and the area change rate is associated with the position offset, it is determined as a real spall.

[0044] Thus, through the chip removal feature image verification step, the accuracy of chip removal detection can be further improved, the false detection rate can be reduced, and the safety and reliability of railway tracks can be ensured. Compared with the prior art, the technical solution of the present application can more effectively distinguish chip removal from static foreign objects, avoiding resource waste and potential safety hazards caused by false detection.

[0045] Furthermore, the present application also proposes to establish a chip removal feature database to store the edge contours, motion trajectories, air pressure parameters, and partition offset data of real chip removals; perform clustering analysis on the chip removal contour features through a convolutional neural network to generate chip removal type labels classified by shape and size, and match them with historical data to optimize the decision threshold.

[0046] The technical solution of the present application mainly involves establishing a chip removal feature database and using a convolutional neural network to perform clustering analysis on the chip removal contour features. By storing the edge contours, motion trajectories, air pressure parameters, and partition offset data of real chip removals, the chip removal features can be systematically managed and analyzed. The convolutional neural network is used to perform clustering analysis on these features to generate chip removal type labels classified by shape and size. By matching with historical data, the decision threshold for chip removal determination can be continuously optimized, improving the accuracy and reliability of detection.

[0047] The establishment of the chip removal feature database can be achieved in various ways. For example, a dedicated data storage system can be designed to store and manage the edge contours, motion trajectories, air pressure parameters, and partition offset data of chip removals. The implementation of the convolutional neural network can be based on existing deep learning frameworks such as TensorFlow or PyTorch. By training the model to identify and classify chip removal features, common clustering algorithms such as K-means can be used for data clustering analysis to classify according to the shape and size of chip removals. By continuously updating and optimizing the decision threshold, the detection system can maintain high detection performance in different environments and conditions.

[0048] By establishing a chip removal feature database and using a convolutional neural network for clustering analysis, the present application can effectively manage and analyze chip removal data, improving the accuracy and reliability of chip removal detection. Compared with the prior art, the present application provides a systematic and intelligent chip removal detection method that can better adapt to complex and changing track environments, reducing the occurrence of false detections and missed detections. Thus, the present application has important practical significance in improving track safety and maintenance efficiency.

[0049] Furthermore, the present application also proposes a secondary verification step for block shedding. After the jet verification execution module is activated in the first block shedding verification step, if the block shedding feature in the verification image does not move, the secondary verification mode is triggered. In the secondary verification mode, the air circuit control system raises the air pressure of the jet group to a preset high-pressure threshold and adjusts the angle of the jet nozzle through the swing assembly for secondary spraying. After the secondary spraying, the secondary verification image is obtained through the image continuous acquisition step. If the block shedding feature still does not move, it is determined as a static foreign object, and this feature is marked as a non-block shedding type. A foreign object database is established to record the static foreign object, which is used to store the position, air pressure parameters, and image features of the static foreign object, and is synchronously updated to the railway operation and maintenance management system to eliminate false detection interference.

[0050] The secondary verification step for block shedding in the present application aims to further improve the accuracy of block shedding detection and solve the possible false detection problems in the first verification. By raising the air pressure of the jet group and adjusting the angle of the jet nozzle for secondary spraying, the block shedding can be more effectively driven to move, so as to more accurately judge whether the block shedding feature is a real block shedding.

[0051] Furthermore, the secondary verification step for block shedding in the present application adopts a higher air pressure and the method of adjusting the angle of the jet nozzle to ensure that it can fully act on the block shedding feature area during the secondary verification. If the block shedding feature still does not move after the secondary verification, it can be more accurately determined as a static foreign object, and this feature is marked as a non-block shedding type. This process not only improves the accuracy of block shedding detection, but also further optimizes the performance of the block shedding detection system by establishing a foreign object database to store the position, air pressure parameters, and image features of the static foreign object.

[0052] Generally speaking, by adding the secondary verification step for block shedding, the present application significantly improves the accuracy and reliability of block shedding detection. Compared with the prior art, the present application can more effectively eliminate false detection interference, ensure the accuracy of the block shedding detection result, and thus improve the efficiency and safety of railway track maintenance.

[0053] Furthermore, the present application also proposes to calculate the pixel area, aspect ratio, and geometric center coordinates of the block shedding feature in the image, and perform spatial mapping in combination with the actual size of the track to output the estimated physical size of the block shedding and the coordinates of the block shedding position.

[0054] During the block shedding detection process, a vision camera is used to obtain the sleeper image and the verification image. The block shedding feature is extracted through the image edge detection algorithm model, and the pixel area, aspect ratio, and geometric center coordinates of the block shedding feature in the image are calculated. Based on the mapping relationship between the actual size of the track and the image resolution, these image features are converted into the estimated physical size and the coordinates of the block shedding position.

[0055] The newly added technical features include calculating the pixel area, aspect ratio, and geometric center coordinates of the chipped feature, and performing spatial mapping in combination with the actual size of the track. These technical features can be achieved in the following ways: First, extract the edge contour of the chipped feature through an image processing algorithm, and then calculate the pixel area enclosed by the contour. Second, calculate the aspect ratio of the chipped feature and determine its geometric center coordinates. Finally, use the mapping relationship between the actual size of the track and the image resolution to convert the above image features into physical size estimation values and the coordinates of the chipped drop position.

[0056] Through the above technical solution, the physical size and drop position of the chipped can be accurately estimated, thereby improving the accuracy and reliability of chipped detection. Compared with the prior art, the solution of this application can provide more accurate chipped positioning and size estimation, which helps to timely detect and repair chipped, and avoid potential safety hazards caused by chipped problems.

[0057] Furthermore, this application also proposes a chipped detection system for the railway track area, including: A track image acquisition module for obtaining sleeper images and verification images through a vision camera; an edge detection and processing module configured to run an image edge detection algorithm model to extract chipped features; a jet verification execution module including symmetrically arranged jet groups, a swing drive unit, and a pressure regulation unit for driving the chipped to move directionally; an image analysis and comparison module for registering images, dividing partitions, and calculating the position offset; a data storage and output module for storing chipped feature data and generating a detection report.

[0058] The technical solution of this system aims to solve the problems of false detection and missed detection in track chipped detection. The track image acquisition module obtains sleeper images and verification images, the edge detection and processing module processes the images to extract chipped features, the jet verification execution module drives the chipped to move directionally through the jet groups, the swing drive unit, and the pressure regulation unit to verify the authenticity of the chipped features, the image analysis and comparison module registers the images, divides partitions, and calculates the position offset of the chipped features, and the data storage and output module is responsible for storing chipped feature data and generating a detection report.

[0059] The newly added technical features of this system include a dynamic air pressure adaptation algorithm, specifically including: a chipped area quantization sub-step, calculating the pixel area S of the chipped feature in the image through the edge detection and processing module, and converting it into a physical area S1 based on the mapping relationship between the actual size of the track and the image resolution; an air pressure calculation sub-step, setting a basic air pressure threshold P0, and dynamically adjusting the first chipped verification injection air pressure value P according to the following formula: , where S2 is a preset reference area threshold, α is the area weight coefficient, and its value range is 0.2 ≤ α ≤ 0.5.

[0060] The edge detection processing module calculates the pixel area of the block loss feature in the image, and converts it into the physical area based on the mapping relationship between the actual size of the track and the image resolution. The air pressure calculation sub-step sets the basic air pressure threshold, and dynamically adjusts the first jet air pressure value for block loss verification according to the formula. In the secondary verification mode, the secondary jet air pressure is 1.2 to 2 times that of the first jet air pressure.

[0061] Furthermore, through the dynamic air pressure adaptation algorithm, the system can dynamically adjust the jet air pressure according to the physical area of the block loss feature, improving the accuracy and reliability of block loss verification. In this way, the system can more effectively detect and verify the block loss features in the track area, reduce the situations of false detection and missed detection, and provide reliable data support for track maintenance.

[0062] Furthermore, this application also proposes a block loss detection system for railway track areas, including: A track image acquisition module for obtaining sleeper images and verification images through a vision camera; an edge detection processing module configured to run an image edge detection algorithm model to extract block loss features; a jet verification execution module including symmetrically arranged jet groups, a swing drive unit, and a air pressure regulation unit for driving the block loss to move directionally; an image analysis and comparison module for registering images, dividing partitions, and calculating the position offset; a data storage and output module for storing block loss feature data and generating a detection report.

[0063] The system obtains sleeper images and verification images through a vision camera, extracts block loss features using the image edge detection algorithm, and drives the block loss to move directionally through the jet verification execution module. The image analysis and comparison module is responsible for registering images, dividing partitions, and calculating the position offset, while the data storage and output module is used to store block loss feature data and generate a detection report.

[0064] Furthermore, the jet verification execution module includes symmetrically arranged jet groups, a swing drive unit, and a air pressure regulation unit. The jet groups are responsible for jetting airflows to the area where the block loss feature is located. The swing drive unit adjusts the angle of the jet nozzle, and the air pressure regulation unit adjusts the pressure of the airflows. Through the coordinated work of these components, the block loss can be effectively driven to move, thereby verifying whether the block loss feature actually exists.

[0065] The advantage of this system is that it can automatically detect and verify the block loss features in the railway track area, reducing the workload and false detection rate of manual inspections. Compared with the prior art, this system combines multiple sensors and algorithms to improve the accuracy and efficiency of detection, especially with stronger adaptability in complex environments. Thus, the track block loss problems can be discovered and processed more timely, ensuring the safety and reliability of railway operations.

Claims

1. A method for detecting falling blocks in a railway track area, characterized in that: The steps include: A track area image acquisition step, in which a track area image of the vehicle is acquired by a visual camera as a sleeper image; A step of detecting a falling block image, extracting falling block features in the sleeper image based on a preset image edge detection algorithm model, wherein the image edge detection algorithm model is configured to integrate Canny edge detection and multi-scale gradient feature extraction; A block drop verification step, when the block drop feature is detected, a block drop verification device is started, wherein the block drop verification device includes an adjustable jet group for jetting air to the area where the block drop feature is located to drive the block drop to move; The image continuous acquisition step is to continuously acquire images of the track area as the verification images after the block drop verification device is started; The step of comparing and outputting the position of the fallen block is to align the sleeper image and the verification image, divide the coordinate system based on the two tracks, and compare the relative position offset of the fallen block features in the two track coordinate systems. If the offset exceeds a preset threshold, it is determined to be a real fallen block and the marking result is output.

2. The method for detecting block drop in a railway track area according to claim 1, characterized in that: The sleeper image and the verification image are processed by a partitioning algorithm, and are divided into a number of parallel partitions with two tracks as the boundary. Each partition is parallel to the track direction and has an adjustable width. Each partition corresponds to an independent position code; When the block drop feature is detected to be located in a certain partition, the jet group of the block drop verification device is oriented and adjusted to match the spatial coordinates of the partition.

3. The method for detecting falling blocks in railway track area according to claim 2, characterized in that: The block drop verification device comprises at least two jet groups, which are respectively deployed on both sides of the track and symmetrically distributed on both sides of the track centerline; Each jet group includes a jet nozzle, a swing assembly for driving the jet nozzle to deflect an angle, and a gas source for providing high-pressure gas to the jet nozzle. The block drop verification device is equipped with a gas path control system; When the block drop feature is located in a partition on one side of the track centerline, the jet group close to this side is started, and the swing component adjusts the jet angle of the jet nozzle according to the partition coordinates, and the jet nozzle drives the airflow to act on the block drop feature area.

4. The method for detecting falling blocks in a railway track area according to claim 3 is characterized in that: The partition width of the verification image is strictly equal to the partition width of the sleeper image, and the partition boundaries are aligned by an image registration algorithm; In the step of comparing and outputting the position of the dropped block, if the offset of the dropped block feature in the verification image relative to the sleeper image spans at least one partition, it is judged as a valid movement, and a dropped block motion trajectory report is generated by combining the air pressure parameters of the jet group and the partition offset distance.

5. The method for detecting falling blocks in railway track area according to claim 4 is characterized in that: It also includes the step of block feature image verification, specifically: The edge contours of the block-dropping features in the sleeper image and the verification image are extracted respectively, and the contour similarity and area change rate of the two are calculated; If the contour similarity is lower than the preset value and the area change rate does not exceed the threshold, it is judged as a static foreign body rather than a falling block; If the contour similarity meets the requirements and the area change rate is associated with the position offset, it is determined to be a real block drop.

6. The method for detecting falling blocks in a railway track area according to claim 5, characterized in that: It also includes the steps of storing and classifying lost blocks, specifically: Establish a block drop feature database to store the edge contour, motion trajectory, air pressure parameters and partition offset data of real block drops; The convolutional neural network is used to perform cluster analysis on the block contour features to generate block type labels classified by shape and size, and match them with historical data to optimize the judgment threshold.

7. The method for detecting block drop in a railway track area according to claim 5, characterized in that: It also includes the secondary verification step of dropped blocks, specifically: When the jet verification execution module in the first block drop verification step is started, if the block drop feature of the verification image does not move, the secondary verification mode is triggered; In the secondary verification mode, the air path control system increases the air pressure of the jet group to a preset high pressure threshold, and adjusts the jet nozzle angle through the swing assembly to perform secondary injection; After the secondary injection, a secondary verification image is obtained through the image continuous acquisition step. If the block-drop feature still does not move, it is determined to be a static foreign body and the feature is marked as a non-block-drop type; A foreign object database is also established to store the location, air pressure parameters and image features of static foreign objects, and is simultaneously updated to the railway operation and maintenance management system to eliminate false detection interference.

8. The method for detecting block drop in a railway track area according to claim 7, characterized in that: The block drop image detection step also includes calculating the pixel area, aspect ratio and geometric center coordinates of the block drop feature in the image, and performing spatial mapping in combination with the actual size of the track to output the estimated physical size of the block drop and the coordinates of the block drop position.

9. The method for detecting falling blocks in a railway track area according to claim 8, characterized in that: The block-dropping image detection step is configured with a dynamic air pressure adaptation algorithm, specifically including: The block area quantification sub-step calculates the pixel area S of the block feature in the image through the edge detection processing module, and converts it into a physical area S1 based on the mapping relationship between the actual track size and the image resolution; In the air pressure calculation sub-step, the basic air pressure threshold value P0 is set, and the first block drop verification injection air pressure value P is dynamically adjusted according to the following formula: ， Among them, S2 is the preset reference area threshold, is the area weight coefficient, and its value range is 0.2 ≤ α ≤ 0.5; In the secondary verification mode, the secondary injection gas pressure is 1.2 to 2 times of p.

10. A railway track area block drop detection system, characterized in that: include: Track image acquisition module, used to obtain sleeper images and verify images through visual cameras; An edge detection processing module is configured to run an image edge detection algorithm model to extract block features; The jet verification execution module includes a symmetrically arranged jet group, a swing drive unit and an air pressure control unit, which is used to directional drive the block to move; Image analysis and comparison module, used to register images, divide partitions and calculate position offsets; The data storage and output module is used to store the block drop feature data and generate the detection report.

Images