A method and system for automatically detecting the state of EMU skirt based on computer vision

By arranging specified color bars on the skirt plates of high-speed rail EMUs and using computer vision technology for image processing, the problems of low loose detection efficiency and high misjudgment rate in the existing technology are solved, and precision is achieved The integrated and automated skirting board status detection has early warning of potential dangers and improved safety guarantees.

CN114549865BActive Publication Date: 2024-11-19CHINA RAILWAY SIYUAN SURVEY & DESIGN GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210176140.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-24
Publication Date
2024-11-19
Estimated Expiration
2042-02-24

AI Technical Summary

Technical Problem

The prior art is inefficient when detecting the looseness of the skirt plate of the high-speed rail EMU, and it is prone to misjudgment and misjudgment. The traditional method has a low detection rate for the looseness of the fastener, making it difficult to warning for potential dangers in advance.

Method used

Using a computer vision-based method, by arranging designated color bars parallel to each other on the skirt board and the side wall of the vehicle body, images are collected and camera calibration, color space transformation, histogram equalization, noise reduction processing and morphological processing are carried out. Identify the color bar and calculate the rotation angle of the skirt plate to measure the angle between the skirt plate and the side wall of the vehicle body to achieve contactless automatic detection.

Benefits of technology

Achieved precision, automated, non-contact angle measurement between the side wall of the skirt and the vehicle body, and early warning of possible risky conditions such as loosening or falling off of the skirt, improving the accuracy and safety of the inspection, reducing the hardware Require.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114549865B_ABST
    Figure CN114549865B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for automatically detecting the state of a train skirt panel based on computer vision. The method comprises: pre-arranging at least two color strips painted with a specified color and parallel to each other on each skirt panel, pre-arranging at least two color strips painted with a specified color and parallel to each other on the side wall of the vehicle body where the skirt panel is installed, and the long sides of the color strips on the skirt panel and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt panel, collecting images of the skirt panel and the side wall of the vehicle body, identifying the color strips on the skirt panel and the side wall of the vehicle body in the image, and calculating the rotation angle of the skirt panel according to the color strip identification result. The present invention can realize non-contact measurement of the angle between the skirt panel and the side wall of the train, thereby providing early warning of dangerous situations such as possible loosening or even falling off of the skirt panel.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic detection, and more specifically, relates to a method for automatically detecting the state of a motor vehicle skirt plate based on computer vision. Background Art

[0002] The skirt of a high-speed train is also called a fully enclosed equipment compartment. It is a baffle arranged on both sides of the lower body of a high-speed train. The skirt of the train can protect the body and equipment, guide the airflow inside and outside the car, and facilitate maintenance personnel to check key equipment. The high-speed operation of the train will generate vibration, and coupled with the influence of other adverse factors, it will cause the skirt to loosen, which not only makes the original function of the skirt lose, but also may endanger the safety of passengers and railway staff on the train. Therefore, the skirt needs to be inspected.

[0003] At present, relevant departments need to arrange special personnel to check the status of each set of skirts in the short time when the EMU is put into storage, which is not only time-consuming and labor-intensive, but also prone to misjudgment and omission, and is inefficient. In addition, although there are some solutions for EMU fault diagnosis using computer vision detection technology, they mostly focus on the identification of foreign objects attached to the car body and the identification of component defects, such as detecting the missing bolts on the skirts. The problem with these solutions is that the frequency of missing parts is actually very small. Usually, fasteners do not fall off completely, but are loose. For skirts, vibration may cause the skirts to loosen, but at this time the fasteners still exist, and the detection rate of traditional detection methods will be greatly reduced. If the change in the rotation state of the skirt around the bottom folding interface can be detected, it is possible to warn of further loosening of the skirt, thereby reducing the possibility of dangerous situations. Summary of the invention

[0004] In response to at least one defect or improvement need in the prior art, the present invention provides a method and system for automatic detection of the status of a train skirt panel based on computer vision, which can realize non-contact measurement of the angle between the skirt panel and the side wall of the train, thereby providing early warning of dangerous situations such as loosening or even falling off of the skirt panel.

[0005] To achieve the above-mentioned purpose, according to the first aspect of the present invention, a method for automatic detection of the state of a train skirt panel based on computer vision is provided, wherein at least two color strips painted with a specified color and parallel to each other are pre-arranged on each skirt panel, and at least two color strips painted with a specified color and parallel to each other are pre-arranged on the side wall of the vehicle body on which the skirt panel is installed, and the long sides of the color strips on the skirt panel and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt panel, images of the skirt panel and the side wall of the vehicle body are collected, the color strips on the skirt panel and the side wall of the vehicle body are identified in the images, and the rotation angle of the skirt panel is calculated according to the color strip recognition results.

[0006] Furthermore, before the color strips on the skirt and the side wall of the vehicle body are identified, the method includes:

[0007] The camera that collects the image is calibrated to obtain the camera's intrinsic parameter matrix and distortion coefficients, and the image is corrected using the camera's intrinsic parameter matrix and distortion coefficients.

[0008] Furthermore, before the identification of the color strip on the skirt panel and the color strip on the side wall of the vehicle body, the method further includes:

[0009] Transform the corrected image into an image in the HSV color space;

[0010] Perform histogram equalization on the image in HSV color space;

[0011] The image after histogram equalization is separated into three HSV channels, and the H channel threshold is selected according to the color of the color bar. The separated image is binarized using the H channel threshold to obtain a binary image.

[0012] Perform noise reduction on the binary image;

[0013] Perform morphological processing on the denoised image.

[0014] Furthermore, the identification of the color strips on the skirt and the side wall of the vehicle body includes:

[0015] Perform contour extraction on the image to obtain multiple point sets corresponding to multiple contours;

[0016] The minimum enclosing rectangle is used to fit the contour of each set of points, and the color strips of the skirt and the side wall of the vehicle body are determined based on the rectangle.

[0017] Furthermore, the color strips of the skirt panel and the side wall of the vehicle body meet the preset area range requirements and aspect ratio range requirements, and the color strips of the skirt panel and the side wall of the vehicle body are determined according to the rectangle, including:

[0018] Calculate the coordinates of the four vertices of all rectangles;

[0019] The length and width of each rectangle are determined according to the vertex coordinates of each rectangle, and only rectangles whose long sides are perpendicular to the apron rotation axis and meet the preset area range requirements and aspect ratio range requirements are retained.

[0020] Further, the calculating the skirt board rotation angle according to the color strip recognition result includes:

[0021] For each identified color bar, solve the equation of the color bar symmetry axis line parallel to the long side of the color bar;

[0022] Identify the equation of the color stripe symmetry axis belonging to the skirt and the equation of the color stripe symmetry axis belonging to the side wall of the vehicle body;

[0023] Solve the vanishing point coordinates of the linear equation of the color stripe symmetry axis belonging to the skirt and the vanishing point coordinates of the linear equation of the color stripe symmetry axis belonging to the side wall of the vehicle body respectively;

[0024] Calculate the scale factors of the skirt and the side wall of the vehicle body according to the vanishing point coordinates and the camera intrinsic parameter matrix;

[0025] The angle between the skirt and the side wall of the vehicle body is calculated according to the scale factor of the skirt and the side wall of the vehicle body, the coordinates of the vanishing point and the camera intrinsic parameter matrix.

[0026] Furthermore, if the number of color stripes on the skirt is not less than three, and the number of color stripes on the side wall of the vehicle body is not less than three, the vanishing point coordinates of the linear equation of the symmetry axis of the color stripes belonging to the skirt include:

[0027] The linear equations of the color strip symmetry axes belonging to the skirt board are combined in pairs to solve the coordinates of multiple vanishing points on the skirt board, and the coordinates of multiple vanishing points on the skirt board are fitted by the least square method to obtain the precise values ​​of the vanishing point coordinates of the skirt board;

[0028] The linear equations of the color strip symmetry axes belonging to the side wall of the vehicle body are combined in pairs to solve the coordinates of multiple vanishing points on the side wall of the vehicle body. The coordinates of multiple vanishing points on the side wall of the vehicle body are fitted by the least square method to obtain the precise values ​​of the vanishing point coordinates of the side wall of the vehicle body.

[0029] Furthermore, the proportional factors of the skirt and the side wall are calculated using the following formula:

[0030]

[0031] Where, (u1, υ1) is the vanishing point coordinate corresponding to the skirt, (u2, υ2) is the vanishing point coordinate corresponding to the side wall of the vehicle body, K is the camera internal parameter matrix, ρ1 and ρ2 are the scale factors of the skirt and the side wall of the vehicle body respectively, and T represents the matrix inversion;

[0032] The angle between the skirt and the side wall of the vehicle body is calculated using the following formula:

[0033]

[0034] Where θ is the angle between the skirt and the side wall.

[0035] Furthermore, CCD cameras are arranged on both sides of the train track to collect images of the train skirt and the side walls of the train body.

[0036] According to a second aspect of the present invention, there is provided a train skirt state automatic detection system based on computer vision, comprising:

[0037] It includes an image acquisition module, which is used to acquire images of the skirt board and the side wall of the vehicle body, wherein at least two color strips painted with a specified color and parallel to each other are pre-arranged on the skirt board, and at least two color strips painted with a specified color and parallel to each other are pre-arranged on the side wall of the vehicle body, and the long sides of the color strips on the skirt board and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt board;

[0038] The data processing module is used to identify the color strips on the skirt and the side wall of the vehicle body in the image, and calculate the rotation angle of the skirt according to the color strip identification result.

[0039] In general, compared with the prior art, the present invention has beneficial effects: it realizes precise, automated, and non-contact measurement of the angle between the skirt and the train side wall, thereby providing early warning of possible loosening or even falling of the skirt, and the method is simple and direct, with low hardware requirements.

[0040] (1) It can obtain more detailed information about the position and posture of the EMU skirt, and can automatically detect subtle changes in the skirt, thereby providing early warning of possible failure of the skirt, thereby improving the safety and fault detection rate of the EMU defect automatic detection system.

[0041] (2) Using color recognition as a target recognition method is more direct than straight line detection, reduces the complexity of detection, avoids interference from irrelevant objects and the operation of screening out a large number of irrelevant objects, and can also improve the accuracy of detection.

[0042] (3) In this scheme, only one color picture is required for each skirt panel, based on which the position information can be calculated, which also reduces the relevant requirements for hardware. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of a train skirt and a car body side wall according to an embodiment of the present invention;

[0044] Figure 2 Schematic diagram of image acquisition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] like Figure 1As shown, an automatic detection method for train skirt panel status based on computer vision according to an embodiment of the present invention comprises: pre-arranging at least two color strips 3 painted with a specified color and parallel to each other on each skirt panel 1; pre-arranging at least two color strips 4 painted with a specified color and parallel to each other on the vehicle body side wall 2 on which the skirt panel is installed; and the long sides of the color strips on the skirt panel and the color strips on the vehicle body side wall are perpendicular to the skirt panel rotation axis 5; collecting images of the skirt panel and the vehicle body side wall, identifying the color strips on the skirt panel and the vehicle body side wall in the image, and calculating the skirt panel rotation angle according to the color strip recognition result.

[0047] The color bar is painted with a specified color, which can be but is not limited to blue. Figure 1 Two skirt panels are shown, one of which is in a closed state and the other is in an open state. Two color strips are pre-arranged on each skirt panel, and two color strips are also pre-arranged on the side wall of the vehicle body near the skirt panel.

[0048] like Figure 2 As shown, images of the train skirt and side walls are collected by arranging CCD cameras on both sides of the train track.

[0049] In another embodiment, a method for automatically detecting the status of a train skirt panel based on computer vision according to an embodiment of the present invention comprises steps (1) to (16).

[0050] (1) Use a CCD industrial camera with a fixed focal length and a camera parameter calibration board to calibrate the camera, thereby obtaining the camera's internal parameter matrix K and its distortion coefficient; paint color strips of specified colors at appropriate locations on the skirt and side walls of the EMU. Assume that there are N strips on a skirt and N strips on the corresponding side walls of the vehicle body.

[0051] When the EMU passes by the CCD industrial cameras arranged on both sides of the track, the cameras continuously obtain side photos of the EMU.

[0052] After acquiring the image, the computer can use the previously obtained camera intrinsic parameter matrix and distortion coefficient to correct the captured image in real time. Usually, when using a CCD camera to image an object, it involves the conversion of information between several coordinate systems. The object's world coordinates are converted to camera coordinates, then to image physical coordinates, and then to image pixel coordinates. Due to the principle of lens imaging, certain distortions are inevitably generated in camera imaging, which requires correction of the captured image.

[0053] (2) Use computer vision related software to transform the image obtained in the previous step from the R (red) G (green) B (blue) color space to the H (hue) S (saturation) V (tone) color space. The H, S, and V values ​​are all functions of the three values ​​of R, G, and B, and there are many options for conversion methods.

[0054] (3) Perform histogram equalization on the result obtained in the previous step. Usually, the components of the dark image histogram are concentrated at the lower grayscale value end, while the components of the bright image histogram tend to be distributed at the higher grayscale value end. If the grayscale histogram of an image almost covers the entire grayscale value range, and the entire grayscale value distribution is approximately uniform, then the image has a larger grayscale dynamic range and higher contrast, and its details will be richer. This processing is to facilitate subsequent recognition.

[0055] (4) The result obtained in the previous step is separated into three channels: H, S, and V. The H channel threshold is selected according to the specific color of the applied functional color strip, and the entire image is binarized to identify the color strips on the skirt and the side wall of the vehicle body. However, interference parts may also be identified at this time, which will be processed later. The selection of the H channel threshold is related to the color of the color strip.

[0056] (5) The results obtained in the previous step are subjected to image preprocessing operations such as median filtering and Gaussian filtering. Usually, the directly obtained images are accompanied by a lot of noise, which may come from hardware device defects, image transmission, improper encoding and decoding, acquisition environment influence, etc. In order to improve the reliability of subsequent recognition, the noise in these images must be processed. Median filtering is often used to process salt and pepper noise in images, and at the same time, it can prevent the signal edge from being blurred by noise or the filtering method itself. Gaussian filtering is often used to eliminate Gaussian noise. The purpose of these operations is to reduce noise.

[0057] (6) The result obtained in the previous step is subjected to morphological processing such as opening, closing and dilation to reduce the interference of irrelevant objects and enhance the fitting accuracy. After the binary image is obtained in the previous step, the color bar (or other irrelevant interference) and the background are black (grayscale value is 0) and white (grayscale value is 255), or vice versa (white and black, both methods are acceptable). In fact, the process performed here is corrosion-dilation-dilation-corrosion-dilation. The opening operation makes the contour of the binary image smoother, disconnects narrow connections, and eliminates burrs, while the closing operation makes the contour smoother, bridges narrow discontinuities, and fills small holes.

[0058] (7) Perform contour extraction on the result obtained in the previous step to obtain multiple sets of points corresponding to multiple contours. In the image, the multiple contours extracted should be multiple sets of points, which reflect the contour topology information of the target object in the binary image.

[0059] (8) For each set of points, the minimum enclosing rectangle is used to fit the contour, and the color strips of the skirt and the side wall of the vehicle body are determined based on the rectangle.

[0060] At this point, there may still be rectangles in the image that do not correspond to the color bars, for example Figure 1The horizontal rectangle above the car body. At this time, an algorithm needs to be designed to filter out the rectangles that do not correspond to the color strips.

[0061] Therefore, the following processing method is preferably adopted: predetermine the area range requirements and aspect ratio range requirements that the color strips of the skirt and the side walls of the vehicle body meet; solve the four vertex coordinates of all rectangles in the image, first group all the vertex coordinates to calculate the distance, because the long sides of the color strips painted with the specified color are perpendicular to the rotation axis of the skirt, the direction of the corresponding rectangle is judged according to the distance between the vertices, so as to filter out the rectangles whose long sides are not perpendicular to the rotation axis of the skirt. It should be noted that due to the operational error when spraying the color strips, it is not strictly vertical here, but within a certain range threshold; then calculate the area and aspect ratio of each remaining rectangle based on the solved vertex coordinates, and compare them with the area range requirements and aspect ratio range requirements of the rectangles corresponding to the color strips in a cyclic comparison, filter out rectangles that are too large and too small, and only retain rectangles that meet the preset area range requirements and aspect ratio range requirements. At this point, the rectangle coordinate information corresponding to all color strips can be reliably obtained.

[0062] (9) For each rectangle obtained in the previous step, the equation of the color bar symmetry axis parallel to its long side is calculated. After this processing, each rectangle in the image corresponds to a symmetry axis straight line equation.

[0063] (10) In the results obtained in the previous step, the equations of the symmetry axes are stored in a certain data structure. Next, the centroid coordinates of each rectangle are calculated and clustered according to their ordinates in the image to achieve classification, that is, the equations of the symmetry axes of all rectangles are divided into two categories: the equations of the symmetry axes of the skirt and the equations of the symmetry axes of the side walls.

[0064] (11) Classify the linear equations classified in the previous step according to their categories, and solve them by combining the lines in the same category. The vanishing point coordinates. The vanishing point refers to the intersection point formed when parallel lines in the real world are imaged in the camera. In other words, the vanishing point can be regarded as an image point formed by a point at infinity.

[0065] (12) In the result of the previous step, there are two types of vanishing points, Then, a more accurate vanishing point coordinate is obtained by fitting each of them through the least square method, and two vanishing point coordinates are obtained, which belong to the skirt and the side wall respectively. If there are two color strips on the skirt and the side wall of the vehicle body, this step is not required, and the unique vanishing point coordinates of each can be obtained directly through the previous step.

[0066] (13) Use the vanishing point coordinates obtained in the previous step and the camera intrinsic parameter matrix K obtained previously to solve the scale factors ρ1 and ρ2 of the skirt and side wall, as shown in formula (1).

[0067]

[0068] Where (u1, υ1) is the vanishing point coordinate corresponding to the skirt, (u2, υ2) is the vanishing point coordinate corresponding to the side wall of the vehicle body, K is the camera intrinsic parameter matrix, ρ1 and ρ2 are the scale factors of the skirt and the side wall of the vehicle body respectively, and T represents the matrix inversion.

[0069] (14) Using the proportional factors ρ1 and ρ2 of the skirt and side wall obtained in the previous step and the vanishing point coordinates (u1, υ1) and (u2, υ2), combined with the camera intrinsic parameter matrix K, the cosine value of the angle between the skirt and the side wall can be obtained, as shown in formula (2).

[0070]

[0071] Where θ is the angle between the skirt and the side wall.

[0072] (15) Using the cosine value of the angle obtained in the previous step, the angle between the skirt and the side wall is calculated using the inverse cosine function, thereby obtaining accurate information on the skirt position and state.

[0073] (16) The angle information obtained in the previous step is stored, summarized and compared, and then combined with the manually set threshold to determine whether the skirt board has become unbearably loose, thereby issuing a warning to relevant personnel through the information system.

[0074] A computer vision-based automatic train skirt state detection system according to an embodiment of the present invention includes:

[0075] It includes an image acquisition module, which is used to acquire images of the skirt board and the side wall of the vehicle body, wherein at least two color strips painted with a specified color and parallel to each other are pre-arranged on the skirt board, and at least two color strips painted with a specified color and parallel to each other are pre-arranged on the side wall of the vehicle body, and the long sides of the color strips on the skirt board and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt board;

[0076] The data processing module is used to identify the color strips on the skirt and the side wall of the vehicle body in the image, and calculate the rotation angle of the skirt according to the color strip identification result.

[0077] The implementation principle and technical effects of the system are similar to the above method and will not be repeated here.

[0078] It must be noted that in any of the above embodiments, the method is not necessarily executed in sequence according to the sequence number. As long as it cannot be inferred from the execution logic that it must be executed in a certain order, it means that it can be executed in any other possible order.

[0079] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatically detecting the state of a train skirt based on computer vision, characterized in that: Arrange at least two color strips painted with a specified color and parallel to each other on each skirt board in advance, and arrange at least two color strips painted with a specified color and parallel to each other on the side wall of the vehicle body where the skirt board is installed in advance, and the long sides of the color strips on the skirt board and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt board, collect images of the skirt board and the side wall of the vehicle body, identify the color strips on the skirt board and the side wall of the vehicle body in the image, and calculate the rotation angle of the skirt board according to the color strip identification result; The method of identifying the color strips on the skirt and the side wall of the vehicle body includes: The camera that collects the image is calibrated to obtain the camera's intrinsic parameter matrix and distortion coefficients, and the image is corrected using the camera's intrinsic parameter matrix and distortion coefficients.

2. The method for automatic detection of train skirt status based on computer vision as claimed in claim 1, characterized in that: Before the color strips on the skirt plate and the color strips on the side wall of the vehicle body are identified, the method further includes: Transform the corrected image into an image in the HSV color space; Perform histogram equalization on the image in HSV color space; The image after histogram equalization is separated into three HSV channels, and the H channel threshold is selected according to the color of the color bar. The separated image is binarized using the H channel threshold to obtain a binary image. Perform noise reduction on the binary image; Perform morphological processing on the denoised image.

3. A method for automatically detecting the state of a train skirt based on computer vision according to any one of claims 1 to 2, characterized in that: The identification of the color strips on the skirt and the side wall of the vehicle body includes: Perform contour extraction on the image to obtain multiple point sets corresponding to multiple contours; The minimum enclosing rectangle is used to fit the contour of each set of points, and the color strips of the skirt and the side wall of the vehicle body are determined based on the rectangle.

4. The method for automatic detection of train skirt state based on computer vision as claimed in claim 3, characterized in that: The color strips of the skirt panel and the side wall of the vehicle body meet the preset area range requirements and the aspect ratio range requirements. The color strips of the skirt panel and the side wall of the vehicle body are determined according to the rectangle, including: Calculate the coordinates of the four vertices of all rectangles; The length and width of each rectangle are determined according to the vertex coordinates of each rectangle, and only rectangles whose long sides are perpendicular to the apron rotation axis and meet the preset area range requirements and aspect ratio range requirements are retained.

5. The method for automatic detection of train skirt state based on computer vision as claimed in claim 1, characterized in that: Calculating the skirt plate rotation angle according to the color strip recognition result comprises: For each identified color bar, solve the equation of the color bar symmetry axis line parallel to the long side of the color bar; Identify the equation of the color stripe symmetry axis belonging to the skirt and the equation of the color stripe symmetry axis belonging to the side wall of the vehicle body; Solve the vanishing point coordinates of the linear equation of the color stripe symmetry axis belonging to the skirt and the vanishing point coordinates of the linear equation of the color stripe symmetry axis belonging to the side wall of the vehicle body respectively; Calculate the scale factors of the skirt and the side wall of the vehicle body according to the vanishing point coordinates and the camera intrinsic parameter matrix; The angle between the skirt and the side wall of the vehicle body is calculated according to the scale factor of the skirt and the side wall of the vehicle body, the coordinates of the vanishing point and the camera intrinsic parameter matrix.

6. The method for automatic detection of train skirt state based on computer vision as claimed in claim 5, characterized in that: If the number of color stripes on the skirt is not less than three, and the number of color stripes on the side wall of the vehicle body is not less than three, the vanishing point coordinates of solving the linear equation of the axis of symmetry of the color stripes belonging to the skirt include: The linear equations of the color strip symmetry axes belonging to the skirt board are combined in pairs to solve the coordinates of multiple vanishing points on the skirt board, and the coordinates of multiple vanishing points on the skirt board are fitted by the least square method to obtain the precise values ​​of the vanishing point coordinates of the skirt board; The linear equations of the color strip symmetry axes belonging to the side wall of the vehicle body are combined in pairs to solve the coordinates of multiple vanishing points on the side wall of the vehicle body. The coordinates of multiple vanishing points on the side wall of the vehicle body are fitted by the least square method to obtain the precise values ​​of the vanishing point coordinates of the side wall of the vehicle body.

7. The method for automatic detection of train skirt state based on computer vision as claimed in claim 5, characterized in that: The scale factor for the skirt and body sidewall is calculated using the following formula: Where (u1, v1) is the vanishing point coordinate corresponding to the skirt, (u2, v2) is the vanishing point coordinate corresponding to the side wall of the vehicle body, K is the camera internal parameter matrix, ρ1 and ρ2 are the scale factors of the skirt and the side wall of the vehicle body respectively, and T represents the matrix inversion; The angle between the skirt and the side wall of the vehicle body is calculated using the following formula: Where θ is the angle between the skirt and the side wall.

8. The method for automatic detection of train skirt state based on computer vision as claimed in claim 1, characterized in that: CCD cameras are arranged on both sides of the train track to capture images of the train skirt and body side walls.

9. A train skirt state automatic detection system based on computer vision, characterized in that: include: It includes an image acquisition module, which is used to acquire images of the skirt board and the side wall of the vehicle body, wherein at least two color strips painted with a specified color and parallel to each other are pre-arranged on the skirt board, and at least two color strips painted with a specified color and parallel to each other are pre-arranged on the side wall of the vehicle body, and the long sides of the color strips on the skirt board and the color strips on the side wall of the vehicle body are perpendicular to the rotation axis of the skirt board; The data processing module is used to identify the color strips on the skirt and the side wall of the vehicle body in the image, and calculate the rotation angle of the skirt according to the color strip identification result.