Track apparent damage inspection trolley and detection method

Abstract

This invention provides a track surface damage inspection trolley, comprising a track inspection trolley body and a working sensor system mounted on the track inspection trolley body. The working sensor system includes an encoder for detecting travel distance, an inertial measurement unit (IMU) for measuring and correcting the coordinates of the track inspection trolley body, a first image acquisition device for acquiring 3D images of the track structure, a second image acquisition device for acquiring 3D images of the surrounding environment, and a third image acquisition device for acquiring 2D images of the track structure. This invention utilizes the image data of the track structure acquired by the track inspection trolley to accurately obtain spatial information of the track structure. Combined with track surface damage detection methods, it achieves qualitative identification and quantitative measurement of visible track defects, automatically providing key information such as the location, type, and value of the defects. This greatly improves detection efficiency and accuracy, has low operating costs, and strong industrial applicability.

Description

Technical Field

[0001] This invention relates to the field of rail transit track inspection and testing technology, specifically to a track surface damage inspection trolley and inspection method. Background Technology

[0002] Railway tracks are a crucial component of railway lines. Over long periods of use, railway tracks are affected by climatic factors such as rain, dew, extreme cold, frost, and sandstorms, as well as the impact and wear from vehicles traveling on them, resulting in various defects such as cracks, deformation, and wear. If these defects are not detected and addressed promptly, they can threaten train safety and directly impact the safety of railway transportation. Therefore, accurate inspection of railway tracks is a vital link in ensuring safe train operation.

[0003] Railway track inspection can promptly detect and eliminate track defects. Through various inspection methods, railway track inspection can identify track defects in a timely manner and take corresponding maintenance measures to ensure the safety and reliability of railway transportation.

[0004] Currently, railway track inspection mainly employs manual inspection and large inspection vehicles with a speed of 80 km / h. However, manual inspection has the following drawbacks: low efficiency; inspection results are affected by personnel, lack objective standards, resulting in large inspection errors and low accuracy; raw data is not recorded in detail or electronically; missed inspections are prone to occur in nighttime operations; and personnel walking on uneven trackbeds pose safety hazards. Using large inspection vehicles also has the following drawbacks: low identification accuracy; high capital investment and cost; and the need for dedicated drivers and operators, leading to high maintenance costs.

[0005] Chinese invention patent application CN 202221231541.1 discloses a track inspection vehicle. The vehicle is equipped with track detection equipment. While the vehicle runs on the rails, the equipment performs track inspection. The principle of the track detection equipment is as follows: during the vehicle's operation, 2D and 3D cameras photograph the track, and the images are transmitted to an industrial control computer for processing. The resulting track inspection results are displayed on a monitor. Simultaneously, an encoder calculates the real-time rotation speed of the vehicle's wheels, and the industrial control computer calculates the vehicle's mileage based on this speed. Combined with an RFID reader, the vehicle's location is determined, yielding the mileage and sleeper number for precise positioning when damaged rails are detected. However, in practical applications, this method does not take into account the vibration of the vehicle body during operation. The vibration transmission through rigid connections can cause significant errors in the measurements of the sensors mounted on the vehicle. Secondly, it does not protect the camera, posing a risk of equipment damage. Furthermore, the inspection vehicle used in this method is large, cumbersome, and difficult to transport. In addition, this method does not consider measuring and correcting the coordinates of the acquisition platform to compensate for measurement errors caused by external vibrations, affecting the accuracy of the detection. The data source is singular, relying solely on 2D and 3D cameras to photograph the track, without considering the acquisition of information about the surrounding environment of the track to obtain overall image information of the track structure, making it difficult to accurately obtain spatial information of the track structure. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art by providing a track surface damage inspection trolley and inspection method, thereby improving inspection efficiency, reducing inspection errors, improving inspection accuracy, and enabling qualitative identification and quantitative measurement of visible track defects.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A track surface damage inspection trolley includes a track inspection trolley body and an operation sensor system installed on the track inspection trolley body;

[0009] The track inspection trolley body includes a body system, a braking system, an axle system, and an electronic control system. The braking system and the axle system together form the bottom running part of the track inspection trolley body. The body system is installed above the braking system and the axle system. The electronic control system is installed inside the body system and includes an industrial computer, which is used to process the data collected by the operation sensor system.

[0010] The operational sensor system includes an encoder for detecting travel mileage, an inertial measurement unit (IMU) for measuring and correcting the coordinates of the track inspection trolley, a first image acquisition device for acquiring 3D images of the track structure, a second image acquisition device for acquiring 3D images of the surrounding environment, and a third image acquisition device for acquiring 2D images of the track structure. The encoder is mounted on the axle system, the IMU is mounted inside the vehicle body system, and the first, second, and third image acquisition devices are all mounted on the vehicle body system.

[0011] Furthermore, it also includes a rain shelter fixed to the top of the track inspection trolley body, the rain shelter completely covering the track inspection trolley body.

[0012] Furthermore, the vehicle body system also includes a shock-absorbing work platform, which is installed at the rear of the vehicle body system; the shock-absorbing work platform includes a platform bracket and a camera bracket installed on the platform bracket, and the platform bracket and the camera bracket are connected by a spring shock absorber; the first image acquisition device, the second image acquisition device, and the third image acquisition device are all installed on the camera bracket, and multiple first image acquisition devices are spaced apart on the camera bracket.

[0013] Furthermore, the axle system also includes a damping shock absorber, which is connected to the vehicle body system.

[0014] This invention also includes a detection method for detecting apparent damage to tracks using the track inspection trolley body described above, comprising:

[0015] S1. Mark the first image acquisition device and acquire 3D image data of the track structure to be detected;

[0016] S2. Extract feature points from the 3D image data to obtain a feature point matrix;

[0017] S3. After obtaining the feature point matrix, each feature point in the feature point matrix is ​​compared with its 8 surrounding feature points. The feature point corresponding to the maximum value among the 9 feature point values ​​is extracted and set as the image key point. This process is repeated to obtain the key point matrix for each 3D image data.

[0018] S4. Perform feature point matching on multiple key point matrices to obtain the mapping matrix required for image stitching and obtain the image stitching result;

[0019] S5. Based on the image stitching results, detect apparent damage to the track.

[0020] Preferably, in step S1, the Tsai method is used to complete the annotation of the first image acquisition device.

[0021] Preferably, in step S2, the scale-invariant feature transformation method is used to extract data feature points. The specific implementation process includes:

[0022] D(x,y,z,σ)=L(x,y,z,k i σ)-L(x, y, z, k j σ);

[0023] Where L(x, y, z, kσ) represents the original image I when the Gaussian sampling scale is k. org (x, y, z) and scale-variable Gaussian kernel I G The convolution of (x, y, z, kσ), i.e., L(x, y, z, kσ) = I org (x, y, z)*I G (x, y, z, kσ); the subscripts i and j represent the row and column coordinates of the pixel in the two-dimensional graphic, and D(x, y, z, σ) represents the point in the feature matrix.

[0024] Preferably, in step S4, the K-nearest neighbor algorithm, which uses three-dimensional Euclidean distance as a metric, is used for feature point matching.

[0025] Preferably, in step S4, a random sampling fitting method is used to obtain the mapping matrix required for image stitching, specifically as follows:

[0026]

[0027] in, The projection matrix of the camera; the first image acquisition device (33) is provided with two sets, wherein, This represents the pixel coordinates in the first image acquisition device of the first group. This indicates the pixel coordinates in the first image acquisition device of the second group.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] (1) The present invention adopts a modular and lightweight design, which can be disassembled and transported separately when not in use or during transportation. Only two operators are needed to achieve rapid assembly and disassembly. At the same time, a rainproof canopy is designed to protect the equipment inside the vehicle and prevent it from being damaged by rain.

[0030] (2) Both the vehicle body and the axle support are equipped with oil-based shock absorbers to effectively reduce the vibration impact of the track on the vehicle body; inside the anti-vibration work platform, shock absorbers are also used between the support frame and the camera mounting bracket to provide a second layer of shock absorption for precision components and improve the accuracy of equipment data acquisition.

[0031] (3) The present invention collects the attitude positioning data of the track inspection trolley body through the inertial measurement unit (IMU) for coordinate correction of the track inspection trolley body, compensates for measurement errors caused by external vibration, and improves measurement accuracy; and provides positioning information to the system by forming an odometer through the encoder and track wheels, so as to accurately locate when defective rails are detected.

[0032] (4) This invention uses a 3D laser scanner to acquire 3D images of the track structure, a lidar to acquire 3D images of the surrounding environment, and a 360-degree panoramic camera to acquire 2D images of the track structure. It can accurately obtain spatial information of the track structure and ensure the reliability and accuracy of the acquired data through comparison of multi-source data. Combined with track surface damage detection methods, it can achieve qualitative identification and quantitative measurement of visible track defects, and automatically provide key information such as the location, type, and value of the defects, which greatly improves detection efficiency and has strong industrial applicability. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the structure of the track inspection vehicle according to an embodiment of the present invention;

[0034] Figure 2 This is a schematic diagram of the body system of the track inspection trolley according to an embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram of the anti-vibration operation platform of the track inspection trolley according to an embodiment of the present invention;

[0036] Figure 4 This is a schematic diagram of the braking system and axle system of the track inspection trolley according to an embodiment of the present invention;

[0037] Figure 5 This is a schematic diagram of the electrical control system and the operation sensor system of the track inspection trolley according to an embodiment of the present invention;

[0038] In the above attached diagrams: 1-Rail inspection trolley body; 2-Rainproof canopy; 3-Work sensor system; 11-Body system; 111-Floor; 112-Anti-vibration work platform; 1121-Platform support; 1122-Camera support; 1123-Spring shock absorber; 113-Electrical control box; 114-Rainproof canopy support; 115-Front railing; 116-Seat base; 117-Rear railing; 12-Brake system; 121-Master brake cylinder; 1211-Brake pedal; 1212- Handbrake mechanism; 1213-Master cylinder; 122-Oil pipe; 123-Brake slave cylinder; 124-Disc brake mechanism; 13-Axle system; 131-Driven front axle; 132-Drive rear axle; 133-Rail wheel; 134-Damping shock absorber; 135-Drive motor; 14-Battery box; 15-Electrical control system; 151-Industrial computer; 152-High voltage box; 153-Motor driver; 154-Three-electric system accessories; 155-Camera driver; 156-Electronic throttle pedal. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0040] like Figure 1 As shown, this embodiment of the invention provides a track inspection trolley, including: a track inspection trolley body 1, a rainproof canopy 2 fixed to the top of the track inspection trolley body 1, and a work sensor system 3 installed at the rear of the track inspection trolley body 1. The rainproof canopy 2 can completely cover the track inspection trolley body 1. The track inspection trolley body 1 includes a body system 11, a braking system 12, an axle system 13, a battery box 14, and an electronic control system 15. The braking system 12 and the axle system 13 together form the bottom running part of the track inspection trolley body 1. The body system 11 is installed above the braking system 12 and the axle system 13. The battery box 14 and the electronic control system 15 are installed inside the body system 11.

[0041] It should be noted that the rain shelter 2 in this embodiment uses ultra-transparent environmentally friendly soft glass to ensure unobstructed visibility within the field of vision; on sunny days, the rain shelter can be rolled up to the top to ensure ventilation and coolness, and on rainy days, the rain shelter can be lowered to completely cover the main body 1 of the track inspection trolley to protect the equipment inside the trolley and prevent it from being damaged by rain.

[0042] like Figure 2As shown, the vehicle body system 11 in this embodiment serves as the mounting platform and support structure for the braking system 12, axle system 13, battery box 14, and electronic control system 15. To ensure ease of transport, the vehicle body system 11 adopts a modular design, including a floor 111, a shock-absorbing work platform 112 mounted on the rear railing 117, an electronic control box 113 mounted inside the floor 111, a canopy bracket 114, a front railing 115 mounted at the head of the floor 111, a seat base 116 mounted inside the floor 111, and a rear railing 117 mounted at the tail of the floor 111. Both the rear railing 117 and the front railing 115 are connected to the canopy bracket 114 via two round rods and secured with bolts or quick-clamping devices.

[0043] like Figure 3 As shown, the anti-vibration work platform 112 includes a platform bracket 1121 vertically mounted on the rear fence 117 and a camera bracket 1122 vertically mounted on the platform bracket 1121. The platform bracket 1121 and the camera bracket 1122 are connected by a spring shock absorber 1123 to ensure that the equipment can reduce the impact of vibration on multiple cameras during operation.

[0044] like Figure 4 As shown, the axle system 13 includes a driven front axle 131, a driven rear axle 132, track wheels 133, damping shock absorbers 134, and a drive motor 135 mounted on the driven rear axle 132. Track wheels 133 are mounted on both sides of the driven front axle 131 and the driven rear axle 132. Both the driven front axle 131 and the driven rear axle 132 are connected to the floor 111 via two damping shock absorbers 134 and are locked with nuts. The damping shock absorbers 134 effectively reduce the vibration experienced by the track inspection trolley body 1 during travel, minimizing the impact of vibration on the measuring device and avoiding significant errors. In addition, in this embodiment, the four track wheels 133 are made of aluminum alloy hubs with polyurethane coating, which prevents delamination and slippage, and uses an LM wear-type tread surface. The tread rim profile meets the requirements of TB / T449-2003.

[0045] The braking system 12 includes a master cylinder 121, oil lines 122, brake slave cylinders 123 mounted on the drive rear axle 132, and a disc brake mechanism 124 mounted on the driven front axle 131. The master cylinder 121 includes a master cylinder 1213, a brake pedal 1211 mounted on the master cylinder 1213, and a handbrake mechanism 1212 mounted on the master cylinder 1213. The disc brake mechanism 124 includes a disc brake bracket and brake pads. Both the master cylinder 121 and the disc brake mechanism 124 are connected to the driven front axle 131 via oil lines 122, and both the driven front axle 131 and the brake slave cylinders 123 are connected to the drive rear axle 132 via oil lines 122.

[0046] like Figure 5As shown, the main functions of the electronic control system 15 consist of an industrial control computer 151 installed in the seat base 116, a high-voltage box 152 installed in the electrical control box 113, a motor driver 153 installed in the electrical control box 113, three-electric system accessories 154 installed in the electrical control box 113, a camera driver 155 installed in the seat base 116, and an electronic throttle pedal 156 installed on the floor 111. The industrial control computer 151 is equipped with data acquisition and monitoring software to process the data collected by the operation sensor system 3. The battery box 14 is a system that provides stable and reliable power supply support for all equipment in the vehicle that requires power. Through the above structure, the battery box 14 provides power to the high-voltage box 152, which in turn provides power to the motor driver 153, and works with the electronic throttle pedal 156 to control the rotation of the drive motor 135, ensuring the forward and backward movement and speed of the vehicle.

[0047] like Figure 4 and Figure 5 As shown, the operation sensor system 3 includes an encoder 31 mounted on the driven front axle 131, an inertial measurement unit (IMU) 32 mounted in the seat base 116, a first image acquisition device for acquiring 3D images of the track structure, a second image acquisition device for acquiring 3D images of the surrounding environment, and a third image acquisition device for acquiring 2D images of the track structure. The first, second, and third image acquisition devices are all mounted on a camera bracket 1122, and multiple first image acquisition devices are spaced apart on the camera bracket 1122.

[0048] Installing encoder 31 in the driven front axle 131 ensures that the driving rear axle 132 can detect the mileage even when coasting after motor braking. Specifically, this embodiment uses a rotary encoder, which is installed on the shaft of the passive speed measuring wheel. The rotation angle of the shaft is detected by the odometer formed by encoder 31 and track wheel 133. Then, the pulse signals emitted by the encoder are collected by specialized hardware and accumulated to obtain the mileage of the track inspection trolley body 1. Dividing the mileage by the collection time gives the speed of the track inspection trolley body 1.

[0049] The inertial measurement unit (IMU) 32 is installed within the seat base 116 and located at the vehicle's center of gravity. Through built-in algorithms, such as inertial navigation algorithms, it outputs the attitude angles, roll angle, heading angle, and three-dimensional acceleration in the xyz orthogonal coordinate system of the track inspection vehicle body 1. This is used for coordinate correction of the track inspection vehicle and to compensate for measurement errors caused by external vibrations. Furthermore, by combining the IMU 32 with an odometer, passive high-precision positioning can be achieved, providing positioning information for the system.

[0050] In this embodiment, the first image acquisition device is a 3D laser scanner 33, the second image acquisition device is a lidar 34, and the third image acquisition device is a 360-degree panoramic camera 35. A total of 4 3D laser scanners 33, 1 lidar 34, and 1 360-degree panoramic camera 35 are provided.

[0051] Among them, the 3D laser scanner 33 projects a laser with a specific coding structure onto the track infrastructure. Multiple CMOS sensors pick up the images, and then the phase and images are decoded and calculated. Using matching technology and the principle of triangulation, the contours of objects such as rails, fasteners, track beds, and turnouts are accurately determined, providing raw data for monitoring the service status of track structures.

[0052] The lidar 34 is used to obtain parameters such as the target's distance, azimuth, altitude, speed, attitude, and shape, and to generate a real-time, high-resolution 3D map or point cloud of the surrounding environment. It is used to obtain accurate spatial information of the overall track structure, analyze the geometric shape and position information of each component of the track structure, and thus identify the service status of each component and the overall track.

[0053] The 360° panoramic camera 35 uses six high-quality sensors to complete image acquisition, sharpening, stitching, tone mapping, and attenuation correction, outputting panoramic 360° images and videos for collecting overall image information of the track structure.

[0054] This invention also provides a method for detecting apparent damage to rails. Four 3D laser scanners 33 installed on the aforementioned rail inspection trolley are used to acquire image data of rail structures such as rails, fasteners, sleepers, and track bed. Based on this image data, apparent damage to the rails is detected. The specific steps are as follows:

[0055] Step 1: Mark multiple 3D laser scanners and collect 3D image data of the track structure to be inspected.

[0056] This embodiment employs the Tsai method, which involves capturing track images to annotate the track using a 3D laser scanner. Since the 3D laser scanner used in this embodiment is a line-scan camera, meaning it only returns one line of data at a time, the calibration data is collected during the movement of the track inspection trolley 1. Based on the calibrated parameters, multiple 3D laser scanners are used to acquire image data of the track structure.

[0057] Step 2: Using the scale-invariant feature transformation method, feature points are extracted from the image data acquired by the 3D laser scanner to obtain a feature point matrix. The specific implementation process is as follows:

[0058] By subtracting the images of adjacent upper and lower layers in each group of the Gaussian pyramid, a Gaussian difference image is obtained, as shown in Equation (1):

[0059] D(x,y,z,σ)=L(x,y,z,k i σ)-L(x, y, z, k j σ) (1)

[0060] Where L(x, y, z, kσ) represents the original image I when the Gaussian sampling scale is k. org (x, y, z) and scale-variable Gaussian kernel I G The convolution of (x, y, z, kσ), i.e., L(x, y, z, kσ) = I org (x, y, z)*I G (x, y, z, kσ); the subscripts i and j represent the row and column coordinates of the pixel in the two-dimensional graphic, and D(x, y, z, σ) represents the point in the obtained feature matrix.

[0061] Step 3: After obtaining the feature point matrix, each feature point in the matrix is ​​compared with its eight surrounding feature points. If the value of this feature point is the maximum value among the nine feature points, then this feature point is considered a keypoint of the image. Using this method, a keypoint matrix can be extracted from the feature point matrix. This process can be repeated to obtain the keypoint matrix for each 3D image data set.

[0062] Step 4: After obtaining a single keypoint matrix, perform feature point matching on multiple keypoint matrices to obtain the mapping matrix required for image stitching, and obtain the image stitching result.

[0063] This embodiment uses the K-nearest neighbor algorithm, which measures three-dimensional Euclidean distance D, to perform feature point matching on multiple key point matrices, as shown in equation (2):

[0064]

[0065] in, This represents the pixel coordinates in the first group of 3D laser scanners. This indicates the pixel coordinates in the second group of 3D laser scanners.

[0066] The mapping matrix required for image stitching is obtained by using a random sampling fitting method, as shown in equation (3):

[0067]

[0068] in, This is a projection matrix used to convert the pixel coordinates in the second set of 3D laser scanners to the pixel coordinates in the first set of 3D laser scanners.

[0069] Step 5: Based on the image stitching results, the industrial control computer 151 uses a deep learning network. First, it extracts the features of the stitched 3D point cloud through two sensory layers with shared parameters to achieve feature alignment. Then, to enhance data correlation and reduce computation, the deep learning network performs dimensionality reduction on the point cloud data that has been feature aligned. Finally, the network uses a pooling layer to extract global variables of the point cloud and completes point feature extraction through a fully connected layer, dividing the point cloud data into the normal track surface and damaged areas to detect apparent track damage.

[0070] It should be noted that apparent track damage can be detected by combining the YOLO network and the Canny edge detection algorithm, or by using the method in this embodiment.

[0071] The track inspection trolley body 1 in this embodiment adopts a lightweight design. The entire vehicle consists of five modules, including a body system 11, a braking system 12, an axle system 13, a battery box 14, and an electronic control system 15. Disassembly is convenient, with each module weighing no more than 70KG. It can be disassembled and transported separately when not in use or during transport, requiring only two operators for rapid assembly and disassembly. The power battery uses automotive-grade lithium batteries with a BMS battery management system; it is equipped with DCAC and OBC to provide 220VAC power and supports national standard slow charging; its driving range is no less than 20km; it is equipped with a high-precision odometer and track gauge measuring instrument; the working platform has shock absorption capabilities to ensure the onboard equipment meets vibration requirements.

[0072] This embodiment utilizes a 3D laser scanner 33 to acquire 3D image data of track structures such as rails, fasteners, ballast, and turnouts; a 360° panoramic camera 35 to acquire 2D image data of track structures; and a lidar 34 to acquire 3D image data of the surrounding environment of the track. Based on the above data, it combines intelligent sensing systems, the Internet of Things, machine learning, and other technologies to process and analyze the data, automatically providing key information such as the location, type, and value of defects. This enables qualitative identification and quantitative measurement of visible track defects, as well as automated identification of structural damage to the track system and assessment of its service status.

[0073] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not limiting. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the scope of protection of the present invention.

[0074] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0075] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A detection method for detecting apparent damage to tracks using a track inspection trolley, characterized in that, The track apparent damage inspection trolley includes a track inspection trolley body (1) and an operation sensor system (3) installed on the track inspection trolley body (1). The track inspection trolley body (1) includes a body system (11), a braking system (12), an axle system (13), and an electronic control system (15). The braking system (12) and the axle system (13) together form the bottom running part of the track inspection trolley body (1). The body system (11) is installed above the braking system (12) and the axle system (13). The electronic control system (15) is installed inside the body system (11). The electronic control system includes an industrial computer (151), which is used to process the data collected by the operation sensor system (3). The operation sensor system (3) includes an encoder (31) for detecting travel distance, an inertial measurement unit (IMU) (32) for measuring and correcting the coordinates of the track inspection trolley body (1), a first image acquisition device (33) for acquiring 3D images of the track structure, a second image acquisition device (34) for acquiring 3D images of the surrounding environment, and a third image acquisition device (35) for acquiring 2D images of the track structure. The encoder (31) is installed on the axle system (13), the inertial measurement unit (IMU) (32) is installed in the body system (11), and the first image acquisition device (33), the second image acquisition device (34), and the third image acquisition device (35) are all installed on the body system (11). The detection method includes: S1. Mark the first image acquisition device (33) and acquire 3D image data of the track structure to be detected; S2. Extract feature points from the 3D image data to obtain a feature point matrix; S3. After obtaining the feature point matrix, each feature point in the feature point matrix is ​​compared with its 8 surrounding feature points. The feature point corresponding to the maximum value among the 9 feature point values ​​is extracted and set as the image key point. This process is repeated to obtain the key point matrix for each 3D image data. S4. Perform feature point matching on multiple key point matrices to obtain the mapping matrix required for image stitching and obtain the image stitching result; S5. Detect apparent damage to the track based on the image stitching results; In step S2, the scale-invariant feature transformation method is used to extract data feature points. The specific implementation process includes: D(x,y,z,σ)=L(x,y,z,k i σ)-L(x,y,z,k j σ)? Where L(x,y,z,kσ) represents the original image I when the Gaussian sampling scale is k. org (x,y,z) and scale-variable Gaussian kernel I G The convolution of (x,y,z,kσ), i.e., L(x,y,z,kσ)=I org (x,y,z)*I G (x,y,z,kσ); the subscripts i and j represent the row and column coordinates of the pixel in the two-dimensional graphic, and D(x,y,z,σ) represents the point in the feature point matrix; In step S4, a random sampling fitting method is used to obtain the mapping matrix required for image stitching, specifically as follows: in, The projection matrix is ​​provided; the first image acquisition device (33) has two sets, wherein, This represents the pixel coordinates in the first image acquisition device (33) of the first group. Indicates the pixel coordinates in the second group of the first image acquisition device (33); In step S5, the specific implementation process for detecting apparent track damage includes: (a) using a deep learning network, extracting features of the image stitching result through two sensory layers with shared parameters to obtain first point cloud data; (b) reducing the dimensionality of the first point cloud data to obtain second point cloud data; (c) using a pooling layer to extract global variables of the second point cloud data, and extracting point features of the second point cloud data through a fully connected layer, dividing the second point cloud data into the normal track surface and the damaged area.

2. The detection method according to claim 1, characterized in that, It also includes a rain shelter (2) fixed on the top of the track inspection trolley body (1), the rain shelter (2) completely covering the track inspection trolley body (1).

3. The detection method according to claim 1, characterized in that, The body system also includes a shock-absorbing work platform (112), which is installed at the rear of the body system (11); The anti-vibration work platform (112) includes a platform support (1121) and a camera bracket (1122) mounted on the platform support (1121). The platform support (1121) and the camera bracket (1122) are connected by a spring shock absorber (1123). The first image acquisition device (33), the second image acquisition device (34), and the third image acquisition device (35) are all mounted on the camera bracket (1122). Multiple first image acquisition devices (33) are spaced apart on the camera bracket (1122).

4. The detection method according to claim 1, characterized in that, The axle system (13) also includes a damping shock absorber (134), which is connected to the body system (11) via the damping shock absorber (134).

5. The detection method according to claim 1, characterized in that, In step S1, the Tsai method is used to annotate the first image acquisition device (33).

6. The detection method according to claim 1, characterized in that, In step S4, the K-nearest neighbor algorithm, which uses three-dimensional Euclidean distance as a metric, is used for feature point matching.

Images