A method for detecting close contact of turnout point rail based on texture image

Abstract

The present invention belongs to the technical field of rail transit defect detection and provides a method for detecting switch point rail adhesion based on texture images. The method comprises: acquiring a texture image of an imaging area and detecting the steel rail and the switch point rail from the image; after detecting the switch point rail, setting m1 sampling windows in the texture image with the steel rail as the axis and perpendicular to the rail extension direction, projecting pixels within the sampling windows along the rail extension direction, sampling the projected curve within the range [-u,u] with the inner edge of the rail as the center on the projection curve, and using a classifier to determine whether there is insufficient switch point adhesion based on the sampling curve; if there is insufficient switch point adhesion, measuring the width of the trough in the sampling curve as the switch point adhesion gap measurement value, and taking the average or maximum value of the m1 switch point adhesion gap measurement values ​​as the final switch point adhesion gap. The present invention can automatically detect insufficient switch point adhesion and quantitatively measure the gap.

Description

[0001] This application is a divisional application of the following Chinese invention patent application, filed May 25, 2023, application number CN2023105956203, invention title: A method and system for detecting close contact between a switch point and a rail, publication number CN116279650A. The applicant filed this divisional application because the original application had a unity issue as noted by the examiner. Technical Field

[0002] The present invention relates to the technical field of rail transit defect detection, and in particular to a turnout point-rail adhesion detection method based on texture images. Background Art

[0003] Railways are the main arteries of the national economy, and maintaining them is crucial for ensuring safe operation. Turnouts, used to guide trains through track changes, are a crucial component of railways. Turnouts contain pointed rails, which normally need to fit snugly against the rails to ensure proper track guidance. Failure to do so can lead to derailment and accidents.

[0004] Currently, manual inspections are primarily used to detect the tightness of the switch rails by inserting steel rulers of varying thicknesses into the gap between the rail and the switch rail. This method suffers from low detection efficiency. To address this issue, the present invention provides an automatic switch switch tightness detection system that automatically determines whether there is insufficient tightness and quantitatively measures the tightness gap. Summary of the Invention

[0005] In order to solve the problems in the background technology, the present invention proposes a method for detecting close fit of a switch rail based on texture images, which can automatically detect insufficient close fit of the switch rail and quantitatively measure the gap.

[0006] A technical solution adopted by the present invention is:

[0007] A turnout point rail adhesion detection system consists of an imaging control module, a visual imaging module, an image processing module, and a carrying platform. The visual imaging module includes two imaging units located directly above the rails on both sides, which image the rails along the movement direction of the carrying platform. The imaging area covers the rails, point rails, and turnouts, and obtains texture images and / or depth images of the imaging area.

[0008] Furthermore, the imaging unit is composed of an area array camera and a surface light source, the area array camera and the surface light source are coaxially illuminated, and the camera optical axis is located directly above the inner edge of the rail to obtain a texture image of the imaging area.

[0009] Furthermore, the imaging unit is a line array camera and a line light source, the line array camera and the line light source are coaxially illuminated, the optical center of the line array camera is located just above the inner edge of the rail, and a texture image of the imaging area is obtained;

[0010] The imaging resolution of the imaging unit along the moving direction of the carrying platform is not less than 1 mm / pixel, and the imaging resolution in the direction perpendicular to the moving direction of the carrying platform is not less than 0.05 mm / pixel.

[0011] Furthermore, the imaging unit is a line structured light 3D camera, which is located directly above the rails to obtain texture images and depth images of the rails and the point rails.

[0012] Furthermore, the imaging unit is a line structured light 3D camera, which is located obliquely above the inner side of the rail. The line structured light plane is perpendicular to the extension direction of the rail. The 3D camera measurement area covers the top and inner side of the rail, and the top and inner side of the point rail. The measurement accuracy of the 3D camera in the x and y directions is not less than 0.1 mm, and the measurement accuracy in the z direction is not less than 0.05 mm.

[0013] Another technical solution adopted in the present invention is:

[0014] A method for detecting close contact of a switch point rail based on texture images comprises the following steps:

[0015] Acquire an image of the imaging area, and detect a first object from the image;

[0016] In the acquired imaging area image, based on the first object, determining whether there is insufficient close contact of the switch rail;

[0017] If there is insufficient close fit of the switch rail, calculate the close fit clearance of the switch rail;

[0018] Wherein, the imaging area includes rails, point rails, and switches;

[0019] The image includes a texture image and / or a depth image.

[0020] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0021] Acquire a texture image of the imaging area and detect the rails and the point rail from the image;

[0022] After detecting the point rail, m1 sampling windows are set in the texture image with the rail as the axis and perpendicular to the rail extension direction. The pixels within the sampling windows are projected along the rail extension direction. On the projection curve, the projection curve is sampled within the range [-u, u] with the inner edge of the rail as the center. Based on the sampled curve, a classifier is used to determine whether there is insufficient adhesion of the point rail.

[0023] The method for determining insufficient close contact of the switch rail using a classifier is:

[0024] Two sample sets were set manually: close fitting and insufficient close fitting. A classifier was trained using pattern recognition method. The trained classifier was used to classify the sampled curves, and the presence of insufficient close fitting of the point rail was determined based on the classification results.

[0025] If there is insufficient tip rail close fit, measure the width of the trough in the sampling curve as the tip rail close fit gap measurement value, and take the average or maximum value of m1 tip rail close fit gap measurement values ​​as the final tip rail close fit gap.

[0026] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0027] Acquire a texture image of the imaging area and detect the rails and the point rail from the image;

[0028] After detecting the point rail, edge extraction is used in the texture image to detect the inner edge of the rail and the edge of the first step of the point rail. m2 sampling lines are set perpendicular to the rail extension direction with the rail as the axis. The width wi from the inner edge of the rail to the bottom edge of the point rail is calculated. When wi>t, it is determined that there is a gap in the point rail.

[0029] When there is a gap, use wi-t=wi' as the gap width, t is the width when the point rail is in close contact with the rail, and take the average or maximum value of m2 gap widths as the point rail close contact gap.

[0030] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0031] Acquire a depth image of the imaging area, detect the rails from the image, find the inner edge of the rails based on their positions, and detect the slide bed using deep learning methods.

[0032] In the detected slide area, a rectangular detection window is set perpendicular to the rail extension direction. Within the detection window, the distance d1 from the inner edge of the rail to the edge of the first step of the switch rail and the distance d2 to the edge of the second step of the switch rail are measured. It is determined whether d1 and d2 exceed the distance d10 and d20 from the inner edge of the rail to the edge of the first step of the switch rail when the switch rail is in close contact with the rail. If any of these exceeds, the switch rail is not in close contact.

[0033] When there is insufficient fit between the point rails, calculate di=(d1-d10+d2-d20) / 2 as the fit gap between the point rails.

[0034] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0035] Pre-collect the turnout depth image and cut it into sections by fastener, so that each image only contains one row of fasteners. Detect the point rail abutment area and the slide bed. Record the pre-collected turnout depth image and the marked areas of the point rail abutment area and slide bed after cutting as the background database.

[0036] Collect the current turnout depth image, and cut the current image based on the previously collected turnout depth image to generate a foreground image that corresponds one-to-one with the background database;

[0037] Obtain the corresponding foreground image and the previously collected turnout depth image, and determine whether the background database contains the point rail contact area. If so, obtain the slide bed plate marking area and the point rail contact area;

[0038] The measurement area is set according to the position of the slide bed. The width from the outer edge of the rail to the bottom edge of the switch rail is measured in the previously acquired turnout depth image and foreground image.

[0039] In the foreground image, determine whether there is a gap between the point rail and the steel rail. If there is a gap, the point rail is not in close contact. If not, determine whether the width from the outer edge of the steel rail to the bottom edge of the point rail in the foreground image increases and exceeds the limit. If there is an increase or exceeds the limit, it is determined that the point rail is not in close contact.

[0040] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0041] Collect rail and point rail depth images, find the rail and point rail from the rail depth images and detect the slide bed;

[0042] A rectangular detection window is set in the detected slide plate area perpendicular to the extension direction of the rail. Within the detection window, n sampling lines are set perpendicular to the extension direction of the rail to obtain n rail and point rail profile curves.

[0043] Among the n rails and the switch rail contour line, perform line segment fitting on the inner edge of the rail to obtain L1, perform line segment fitting on the vertical edge of the first step of the switch rail to obtain L2, and perform line segment fitting on the vertical edge of the second step of the switch rail to obtain L3;

[0044] Calculate the distance d3 from L1 to L2 and the distance d4 from L1 to L3, and determine whether d3 or d4 exceeds the distance d30 from L1 to L2 and the distance d40 from L1 to L3 when the switch rail is tightly attached to the rail. If either exceeds the distance d3 or d4, then insufficient contact between the switch rail and the rail exists.

[0045] When there is insufficient close contact between the point rail, Di = (d3-d30+d4-d40) / 2 is calculated as the measurement value of the close contact gap between the point rails, and the average or maximum value of the n close contact gap measurement values ​​Di is taken as the close contact gap between the point rails in the detection window.

[0046] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0047] Collect rail and point rail depth images, find the rail and point rail from the rail depth images and detect the slide bed;

[0048] In the slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail. Within the detection window, m3 sampling lines are set perpendicular to the extension direction of the rail to obtain the contour curves of m3 rails and point rails.

[0049] According to the position of the inner edge of the rail, fit the inner vertical edge line segment L4 of the rail; according to the position of the point rail, calculate the fitting circle center O1 of the point rail waist curve, and calculate the distance f1 from the circle center O1 to L4;

[0050] For m3 contour curves, obtain m3 measurement results f1, and determine whether f1 exceeds the distance e1 when the switch rail is close to the rail. If so, the switch rail is not close enough.

[0051] When there is insufficient close contact between the point rail, hi=f1-e1 is calculated as the measurement value of the close contact gap between the point rails; the average or maximum value of the m3 close contact gap measurement values ​​hi is taken as the close contact gap between the point rails in the detection window.

[0052] Furthermore, a method for detecting close contact of a switch point rail comprises the following steps:

[0053] Collect rail and point rail depth images, find the rails and point rails from the rail depth images and the point rail depth images, and detect the slide plate area;

[0054] In the slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail. Within the detection window, the vertical edge plane P1 of the inner side of the rail is fitted according to the position of the inner edge of the rail.

[0055] In the detection window, r sampling lines are set perpendicular to the extension direction of the rail to obtain the contour curves of r rails and the point rail;

[0056] According to the position of the switch rail, calculate the center O2 of the switch rail waist curve fitting circle and calculate the distance si from the center O2 to P1; determine whether si exceeds the distance e2 when the switch rail is close to the rail. If so, it is determined that the switch rail is not close enough;

[0057] When there is insufficient close contact between the switch rails, gi = si - e2 is calculated as the switch rail gap measurement value, and the average or maximum value of the r switch rail close contact gap measurement values ​​gi is taken as the switch rail close contact gap measurement result within the detection window.

[0058] The beneficial effects of the present invention are:

[0059] This invention proposes an automated inspection system and method based on visual inspection methods. By improving the imaging unit within the visual imaging module, it can automatically detect insufficient rail contact and quantitatively measure the gap. When the imaging unit is an area array camera and a surface light source, coaxial lighting and positioning the camera's optical axis directly above the inner edge of the rail enable the acquisition of high-quality, clear images of the gap between the rail and the rail. When the imaging unit is a linear array camera and a line light source, this not only reduces costs but also increases the gap measurement resolution, reaching 0.05 mm / pixel, significantly improving gap measurement accuracy.

[0060] The imaging unit is a line structured light 3D camera that can simultaneously acquire texture images and depth images of the rails and point rails. Based on the texture images and depth images, it can detect and measure insufficient rail contact from multiple dimensions. Furthermore, when the imaging unit is a line structured light 3D camera, and the 3D camera is installed diagonally above the inner side of the rail, the detection system tilts the imaging. When measuring from the inside, it can clearly image the inner edge of the rail. Compared with vertical imaging (the 3D camera is located directly above the rail), the acquired data on the inner edge of the rail is richer, and the distance between the inner edge plane of the rail and the edge plane of the first step of the point rail is used to determine whether it is closely attached, which has better robustness. Moreover, this hardware architecture can also be used for rail profile measurement, achieving the effect of one machine for multiple uses.

[0061] To meet the needs of close-fitting gap measurement, in addition to edge detection and distance measurement through texture images, we also determine the close-fitting situation through the statistics of the measurement results of multiple sampling curves, which has better robustness.

[0062] Furthermore, when using an inclined installation of the 3D camera, the straight line / plane at the edge of the pointed rail, the pointed rail arc and other areas are used to fit the center of the straight line / plane and arc, and the distance from the center of the circle to the straight line / plane is calculated, which has higher measurement accuracy than directly measuring the edge distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The following is a schematic diagram of the system composition;

[0064] Figure 2 This is a schematic diagram of camera installation in Example 1;

[0065] Figure 3 Schematic diagram of rails and point rails;

[0066] Figure 4 This is a schematic diagram of camera installation in Example 5;

[0067] Figure 5 This is a schematic diagram of the close fitting of the point rail;

[0068] Figure 6 This is a schematic diagram of non-close fitting of the point rail;

[0069] 1-rail, 2-visual imaging module, 3-imaging control module, 4-image processing module, 5-carrying platform, 6-area array camera, 7-surface light source, 8-optical axis of area array camera, 9-top surface of rail, 10-inner edge of rail, 11-point rail, 12-sampling window, 14-edge of the first step of point rail, 15-bottom plane of the first step of point rail, 16-edge of the second step of point rail, 17-waist of point rail, 18-bottom of the second step of point rail. DETAILED DESCRIPTION

[0070] The present invention will be described in detail below with reference to the accompanying drawings and embodiments thereof, but the protection scope of the present invention is not limited to the scope of the embodiments described herein.

[0071] like Figure 5 As shown in the figure, the rails and the point rails in the turnout are tightly attached to each other and there is no gap. Figure 6 As shown, there is a gap between the rails and the point rails in the turnout, and they cannot fit tightly together.

[0072] The system for detecting the close contact of switch point rails in the present invention is composed of an imaging control module, a visual imaging module, an image processing module and a carrying platform;

[0073] The imaging control module, image processing module and carrier platform used in the embodiments of the present invention are solutions in the prior art. For example, the imaging control module includes a carrier platform speed measurement unit and an imaging control signal generator. The speed measurement unit accurately measures the speed of the carrier platform; the imaging control signal generator generates an imaging control pulse signal to the visual imaging module based on the imaging resolution requirements of the moving platform's movement speed and movement direction. The image processing module is connected to the visual imaging module, receives the turnout texture image collected by the visual imaging module, and executes the turnout point rail close fit detection algorithm to complete the turnout point rail close fit detection. The carrier platform is a train or electric passenger car or inspection vehicle or patrol robot or hand-pushed trolley, which provides power supply and installation support for the detection system.

[0074] The improvement of the present invention lies in the improvement of the visual imaging module, which includes two imaging units located directly above the rails on both sides, and images the rails along the movement direction of the carrying platform. The imaging area covers the rails, point rails, and switches, and obtains texture images and / or depth images of the imaging area.

[0075] The layout of the visual imaging module varies according to the collected images, and the corresponding detection methods are also different. For specific layouts and detection methods, please refer to the following specific embodiments.

[0076] Example 1

[0077] The switch rail close contact detection system of this embodiment is as follows: Figure 1 As shown, it consists of an imaging control module 3, a visual imaging module 2, an image processing module 4 and a carrying platform 5;

[0078] The imaging control module 3 includes: a carrier platform speed measurement unit and an imaging control signal generator. The speed measurement unit accurately measures the speed of the carrier platform. The speed measurement unit is a wheel speed measurement module based on an encoder, a radar speed measurement module, or an LDV speed measurement module. The imaging control signal generator generates an imaging control pulse signal to the visual imaging module based on the imaging resolution requirements of the moving platform's moving speed and moving direction.

[0079] The visual imaging module 2 includes two imaging units located directly above the rails on both sides, which image the rails along the movement direction of the carrying platform. The imaging area covers the rails, point rails, and switches, and obtains texture images and / or depth images of the imaging area.

[0080] like Figure 2 As shown, the imaging unit includes an area array camera 6 and a surface light source 7; the area array camera 6 and the surface light source 7 are coaxially illuminated, and the optical axis 8 of the area array camera 6 is located directly above the inner edge 10 of the rail 1. The imaging unit is configured to obtain a high-quality, clear image of the gap between the rail and the switch rail.

[0081] The image processing module 4 is connected to the visual imaging module 2, receives the turnout texture image collected by the visual imaging module 2, and executes the turnout point rail close fit detection algorithm to complete the turnout point rail close fit detection;

[0082] The carrying platform 5 is a train, an electric passenger car, an inspection vehicle, an inspection robot, or a hand-pushed trolley, which provides power supply and installation support for the inspection system.

[0083] The turnout rail close contact detection method of this embodiment is as follows:

[0084] S1: Collect rail texture images and detect rail 1 using edge extraction method; detect point rail 11 using point rail bolt or slide plate information;

[0085] S2: After the switch rail 11 is detected, m1 sampling windows 12 are set with the rail 1 as the axis and perpendicular to the rail extension direction. The pixels within the sampling windows are projected along the rail extension direction. On the projection curve, with the inner edge of the rail as the center, the projection curve is sampled within the range [-u, u]. Two sample sets are artificially set: close contact and insufficient contact. A classifier is trained using a pattern recognition method. The sampled curves are classified by the classifier to determine whether there is insufficient contact between the switch rail and the sampled curves.

[0086] S3: When the tip rail is not close enough, measure the width of the trough in the sampled projection curve as the tip rail close gap measurement value; take the average of m1 tip rail close gap measurement values ​​(d1, ..., dm) as the final close gap;

[0087] The imaging unit has been calibrated and can obtain the physical dimensions of the rail plane through image pixels.

[0088] It should be noted that the sampling window projection in S2 is designed to obtain high-quality curve data of the gap between the rail and the switch rail, which can reflect the degree of fit. Training a classifier using two artificially constructed sample sets of close fit and non-close fit, and determining whether fit is achieved through classification, is easier to implement than directly measuring the gap. Using the trough width of the projected curve as the measurement of the gap in close fit in S3 is a simple and highly accurate measurement method.

[0089] Example 2

[0090] The detection system of this embodiment differs from that of Example 1 in that the imaging unit 2 comprises a linear array camera and a linear light source, the linear array camera and the linear light source are coaxially illuminated, the optical center of the linear array camera is located above the inner edge of the rail, the imaging resolution along the direction of movement of the carrying platform is not less than 1 mm / pixel, and the imaging resolution in the direction perpendicular to the direction of movement of the carrying platform is not less than 0.05 mm / pixel.

[0091] This embodiment uses a line array camera in addition to an area array camera and an area array light source, which not only reduces costs but also increases the gap measurement resolution. The resolution of this embodiment reaches 0.05mm / pixel, significantly improving the gap measurement accuracy.

[0092] The processing steps of the turnout rail close contact detection algorithm of this embodiment are as follows:

[0093] S1: Collect rail texture images and detect rails using edge extraction methods; detect the point rails using point rail bolts or slide plate information;

[0094] S2: After the switch rail is detected, edge extraction is used within the rail texture image to detect the inner edge of the rail and the edge of the first step of the switch rail. m2 sampling lines are set perpendicular to the rail extension direction, centered on the rail. The width wi from the inner edge of the rail to the bottom edge of the switch rail is calculated. When wi > t, a gap between the switch rail and the rail is determined. If a gap exists, the gap width is calculated as wi - t = wi', where t is the width of the switch rail when it is in close contact with the rail. The average or maximum value of the m2 gap widths is taken as the switch rail close contact gap measurement value.

[0095] like Figure 3As shown, imaging calibration is performed on the top surface 9 of the rail and the bottom plane 15 of the first step of the point rail 11. When calculating the actual physical distance, the position of the rail edge takes the rail plane calibration parameters, and the position of the edge of the first step of the point rail takes the calibration parameters of the bottom plane of the first step of the point rail.

[0096] Example 3

[0097] The difference between the detection system of this embodiment and Example 2 is that the imaging unit is a line structured light 3D camera, which is located directly above the rail and scans and images the rail and point rail area to obtain texture images of the rail and point rail; the texture image processing method in Example 2 is adopted, and edge detection is used for point rail close detection and close gap measurement.

[0098] Example 4

[0099] The difference of the detection system embodiment 3 of this embodiment is that the rail and the point rail area are scanned and imaged to obtain a depth image of the rail and the point rail; and the close contact detection is performed on the depth image. The specific steps are as follows:

[0100] S1: Collect rail depth images, find rails based on rail height information using threshold segmentation, find the inner edge of the rail based on the rail position, and find the slide bed using deep learning methods;

[0101] S2: In the detected slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail;

[0102] S3: Within the detection window, using projection and edge extraction methods, measure the distance d1 from the inner edge 10 of the rail to the first step edge 14 of the switch rail, and the distance d2 from the inner edge 10 of the rail to the second step edge 16 of the switch rail. Determine whether d1 and d2 exceed the distance d10 and d20, respectively, when the switch rail is in close contact with the rail. If either exceeds the value, insufficient contact exists between the switch rail and the inner edge 10 of the rail. S4: If insufficient contact exists between the switch rail and the inner edge 10 of the rail, calculate di = (d1 - d10 + d2 - d20) / 2 as the switch rail contact gap.

[0103] S5: Perform the above processing on multiple slide bed areas to obtain the point rail close contact state and close contact gap results at each slide bed.

[0104] Example 5

[0105] The imaging unit is the same as that of Example 4, except that the method for detecting the close contact of the switch rail is:

[0106] S1: Pre-collect the turnout depth image and slice it by fastener so that each image contains only one row of fasteners. Use deep learning methods to detect the point rail contact area and the slide bed. Record the sliced ​​depth image and the marked area of ​​the point rail contact area and the slide bed as the background database.

[0107] S2: Collect the depth image to be detected, and cut the current collected image with the background image stored in the background database as the reference to generate a foreground image that corresponds one-to-one with the background database;

[0108] S3: Obtain the corresponding foreground and pre-collected turnout depth images, determine whether there is a switch rail abutment mark in the background database, and if so, obtain the slide bed plate marking area at the marked switch rail;

[0109] S4: According to the position of the slide bed, the measurement area is set, and the width from the outer edge of the rail to the bottom edge of the switch rail is measured in the previously collected turnout depth image and foreground image respectively;

[0110] S5: In the foreground image, determine whether there is a gap between the point rail and the steel rail through threshold segmentation. If there is a gap, the point rail is not closely attached. If not, determine whether the width from the outer edge of the steel rail to the bottom edge of the point rail in the foreground image increases and exceeds the limit. If there is an increase or exceeds the limit, it is determined that the point rail is not closely attached.

[0111] Example 6

[0112] The difference between the detection system of this embodiment and that of embodiment 4 is that the 3D camera is installed obliquely above the inner side of the rail. Figure 4 As shown, the line structured light plane is perpendicular to the rail extension direction, and the 3D camera measurement area covers the top and inner side of the rail, and the top and inner side of the switch rail. The 3D camera's measurement accuracy in the x and y directions is no less than 0.1 mm, and the measurement accuracy in the z direction is no less than 0.05 mm. The processing steps of the switch switch rail close fit detection algorithm are as follows:

[0113] S1: Collect rail and switch rail depth images, find the rail and switch rail areas from the rail depth images and the switch rail depth images respectively; find the slide plate area through deep learning methods;

[0114] S2: In the slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail;

[0115] S3: Within the detection window, set n sampling lines perpendicular to the rail extension direction to obtain n rail and switch rail profile curves; among the n rail and switch rail profile curves, perform line segment fitting on the inner edge of the rail to obtain L1; perform line segment fitting on the vertical edge of the first step of the switch rail to obtain L2, and perform line segment fitting on the vertical edge of the second step of the switch rail to obtain L3;

[0116] S4: Calculate the distance d3 from L1 to L2 and the distance d4 from L1 to L3, and determine whether d3 and d4 exceed the distance d30 from L1 to L2 and the distance d40 from L1 to L3 when the point rail is in close contact. If any of them exceeds the limit, it is determined that there is insufficient close contact, and Di = (d3-d30+d4-d40) / 2 is calculated as the point rail close contact gap measurement value; take the average or maximum value of the n point rail close contact gap measurement values ​​Di as the point rail close contact gap measurement result in the detection window.

[0117] This embodiment uses oblique imaging to clearly image the inner edge of the rail when measuring from the inside. Compared to vertical imaging, this captures richer data on the inner edge of the rail. A line segment can be fitted to the inner edge of the rail, and then the center of a circle can be fitted based on the curve of the first step of the switch rail to calculate the distance from the center of the circle to the inner edge of the rail. Because there is only one piece of data for the inner edge of the rail, it is susceptible to noise interference. In this embodiment, when shooting obliquely, the inner edge of the rail is represented by a line segment, allowing for straight line fitting. This straight line fitting is more resistant to noise interference than using only one piece of data. Therefore, this oblique imaging method can improve the accuracy of measuring the distance from the inner edge of the rail to the switch rail.

[0118] Example 7

[0119] The imaging unit is the same as that of Example 6, except that the processing steps of the switch point rail close contact detection algorithm are as follows:

[0120] S1: Collect the depth image of the rail and the switch rail, find the rail and switch rail area in the depth image, and find the slide bed through the deep learning method;

[0121] S2: In the slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail;

[0122] S3: Within the detection window, set m3 sampling lines perpendicular to the rail extension direction to obtain m3 rail and switch rail profile curves; fit the rail inner vertical edge line segment L4 according to the rail inner edge position; calculate the switch rail waist curve fitting circle center O1 according to the switch rail position, and calculate the distance f1 from the circle center O1 to L4;

[0123] S4: For m3 contour curves, obtain m3 measurement results f1, determine whether f1 exceeds the distance e1 when the point rail is close to the rail, if it exceeds, it is determined that there is insufficient close contact, and calculate hi=fi-e1 as the point rail close contact gap measurement value; take the average or maximum value of the m3 point rail close contact gap measurement values ​​hi as the point rail close contact gap measurement result in the detection window.

[0124] It should be noted that higher accuracy can be obtained by fitting the circle center O1 in S3.

[0125] Example 8

[0126] The imaging unit is the same as that of Example 6, except that the processing steps of the switch point rail close contact detection algorithm are as follows:

[0127] S1: Collect the depth image of the rail and the switch rail, find the rail and switch rail area in the depth image; find the slide bed plate through the deep learning method;

[0128] S2: In the slide bed area, a rectangular detection window is set perpendicular to the extension direction of the rail;

[0129] S3: Fitting the inner vertical edge plane P1 of the rail according to the inner edge position of the rail within the detection window;

[0130] S4: Within the detection window, set r sampling lines perpendicular to the rail extension direction to obtain the profile curves of r rails and the switch rail; calculate the center O2 of the switch rail waist curve fitting circle based on the switch rail position, and calculate the distance si from the center O2 to P1; determine whether si exceeds the distance e2 when the switch rail is close to the rail; if so, determine that there is insufficient close contact, and calculate gi = si - e2 as the switch rail close contact gap measurement value;

[0131] S5: Take the average or maximum value of the n measurement results as the measurement result of the sharp rail close clearance in the detection window.

[0132] This embodiment adopts plane fitting, which can further improve the ability to resist noise interference compared with the straight line segment fitting in Examples 3-5.

[0133] Although the principles of the present invention have been described in detail above in conjunction with the preferred embodiments of the present invention, those skilled in the art should understand that the above embodiments are merely illustrative of the present invention and are not intended to limit the scope of the present invention. The details in the embodiments do not constitute a limitation on the scope of the present invention. Without departing from the spirit and scope of the present invention, any obvious changes such as equivalent transformations and simple substitutions based on the technical solution of the present invention fall within the scope of protection of the present invention.

Claims

1. A method for detecting close contact of switch point rails based on texture images, characterized in that: The turnout point rail close contact detection system is used for detection, which includes the following steps: Acquire a texture image of the imaging area and detect the rails and the point rail from the image; After detecting the point rail, m1 sampling windows are set in the texture image with the rail as the axis and perpendicular to the rail extension direction. The pixels within the sampling windows are projected along the rail extension direction. On the projection curve, the projection curve is sampled within the range [-u, u] with the inner edge of the rail as the center. Based on the sampled curve, a classifier is used to determine whether there is insufficient adhesion of the point rail. If there is insufficient close contact between the switch rails, measure the width of the trough in the sampling curve as the switch rail close contact gap measurement value, and take the average or maximum value of m1 switch rail close contact gap measurement values ​​as the final switch rail close contact gap; The method for determining insufficient close contact of the switch rail using a classifier is: Two sample sets were manually set: close fit and insufficient close fit. A classifier was trained using a pattern recognition method. The trained classifier was used to classify the sampled curves, and the presence of insufficient close fit was determined based on the classification results. The switch point rail close contact detection system consists of an imaging control module, a visual imaging module, an image processing module, and a carrying platform. The visual imaging module includes two imaging units located directly above the rails on both sides, which image the rails along the movement direction of the carrying platform. The imaging area covers the rails, the point rails, and the switches, and obtains a texture image of the imaging area. The imaging unit is composed of an area array camera and a surface light source. The area array camera and the surface light source are coaxially illuminated, and the camera optical axis is located just above the inner edge of the rail to obtain a texture image of the imaging area.

Images