Rail train wheel surface scratch detection method and system

Abstract

The invention provides a rail train wheel surface scratch detection method and system, and the method comprises the steps: obtaining a wheel surface image: obtaining a wheel surface image, collected by a visual camera under the irradiation of a stroboscope, of a wheel surface under high-speed movement; an abnormal feature mapping step: performing change detection on the wheel surface image and the template image at different angles to obtain an abnormal feature group, positioning and splicing the abnormal feature group to obtain an abnormal feature block, and mapping the abnormal feature block in the front image of the wheel surface as a wheel surface abnormal image; a feature depth analysis step: obtaining depth information of a plurality of item points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, the plurality of item points including edge item points and non-edge item points of the abnormal feature block, and comparing the depth information of the edge item points with the depth information of the non-edge item points; an exception screening and matching step: outputting a scratch type or other types according to a comparison result; the method has the advantages that the abrasion of the wheel side face or the shaft surface can be accurately positioned, and the abrasion is accurately judged to be a scratch type.

Description

Technical Field

[0001] The present invention relates to the technical field of rail vehicle wheel tread detection, and more particularly to a rail vehicle wheel tread scratch detection method and system. Background Art

[0002] The wheelset of a rail vehicle is a key component of the train running gear. Its condition directly affects the safety and stability of the train operation. If the wear, cracks, deformation and other defects of the wheelset are not detected and handled in time, serious safety accidents may occur. Therefore, effective inspection of the train wheelset is an important measure to ensure the safety of train operation.

[0003] With the rapid development of high-speed railways, the operating speed of trains is constantly increasing, and the performance requirements for wheelsets are becoming higher and higher. High-speed railway trains need to withstand greater loads and impacts during operation, so the scratches on the wheel surface and wheels will become more serious, resulting in reduced stability of the train during driving, shaking or uneven force, etc.

[0004] In the automated detection of wheel treads for rail vehicles, there are currently some technical solutions for wheel tread detection, such as the Chinese patent "CN110276297B Train Tread Wear Detection System and Method", which provides a thermal imaging image of the bottom of the train collected by thermal imaging, and the tread wear and scratch abnormalities of the wheelset and the track are obtained based on the matching of the thermal imaging image with the model. However, this detection method cannot determine abnormal defects on the wheel surface. When scratches or stains appear on the wheel surface, thermal imaging cannot be identified, so it is difficult to meet the surface scratch detection of the wheel surface. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for detecting scratches on the wheel surface of a railway train, which can accurately locate the wear on the side of the wheel or the surface of the axle and accurately determine whether the wear is a scratch type.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for detecting scratches on a railway train wheel surface comprises the following steps: A wheel surface image acquisition step, in which a visual camera under the illumination of a stroboscope collects an image of the wheel surface under high-speed motion as the wheel surface image; Abnormal feature mapping step, obtaining an abnormal feature group by performing change detection on the wheel surface image and the template image at different angles, then locating and splicing the abnormal feature group to obtain an abnormal feature block, and mapping the abnormal feature block in the wheel surface front image as a wheel surface abnormality map; A feature depth analysis step, obtaining depth information of a plurality of points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, wherein the plurality of points include edge points and non-edge points of the abnormal feature block, and comparing the depth information of the edge points and the non-edge points; In the abnormal screening and matching step, when the depth of the edge item point is less than the depth of the non-edge item point, the output abnormal type is the scratch type, otherwise, the output abnormal type is other types.

[0007] Furthermore, the depth analysis strategy includes an edge trajectory planning step and a depth estimation step. The edge trajectory planning step is to draw a plurality of edge points of the abnormal feature block in the front image of the wheel surface to obtain an edge trajectory line; The depth estimation step selects any point in the edge trajectory as an edge point, selects any point in the area surrounded by the edge trajectory as a non-edge point, calculates the disparity in the wheel surface images at any two different angles, and then obtains the point distance between the item to be measured and the visual camera through depth calculation based on the disparity, and subtracts the point distance from the distance of the visual camera from the wheel surface to obtain the depth value of the item to be measured.

[0008] Furthermore, in the abnormal feature mapping step, edge contours of abnormal features are extracted according to edge detection, the edge contours of each abnormal feature in the abnormal feature group are point matched to obtain a matching point group, and then a transformation matrix is ​​calculated according to the matching point group, one of the abnormal features is aligned with another abnormal feature through the transformation matrix, and after alignment, the two are fused to obtain an abnormal feature block.

[0009] Furthermore, in the abnormal feature mapping step, a transformation matrix is ​​calculated for any wheel surface image and the wheel surface front image, and several edge contour points are selected in the wheel surface image and transformed through the transformation matrix to obtain the position points of the edge contour points in the wheel surface front image, and then the abnormal feature block is matched with the corresponding points used for mapping through the position points, so that the abnormal feature block is mapped as a whole to the wheel surface front image.

[0010] Furthermore, the method also includes a stroboscopic adjustment step, in which the corresponding stroboscope flashing frequency is indexed in a preset table according to the train speed as the current frequency value.

[0011] Furthermore, it also includes a feature verification and analysis step, in which the wheel surface abnormality map is binarized to obtain a binary image, and in the binary image, it is judged based on pixels whether the pixel value in the abnormal feature block is the same as the wheel surface pixel value outside the abnormal feature block. If so, the output abnormality type is a scratch type, if not, the output abnormality type is a stain diffusion type.

[0012] Furthermore, it also includes a feature verification and analysis step. When the pixel value within the abnormal feature block in the binary image is different from the wheel surface pixel value outside the abnormal feature block, the contour line of the abnormal feature block is obtained. If there is a sub-contour line in the contour area of ​​the abnormal feature block, the scratch filling type is output, and the scratch filling type reflects the diffusion of stain filling after the wheel surface is scratched.

[0013] Furthermore, it also includes an instruction feedback step, comparing the depth value with a preset value N1, if the depth value is greater than the preset value N1, outputting a crisis instruction, if the depth value is less than or equal to the preset value N1, outputting a correction instruction.

[0014] Furthermore, the method further includes a scratch area calculation step, obtaining an abnormal feature block under the scratch type, performing contour segmentation on the abnormal feature block to obtain a number of calculable scratch areas, respectively obtaining area values ​​of the scratch areas by area calculation, and accumulating the values ​​to obtain the scratch area value; The instruction feedback step compares the scratch area value with a preset value N2, and outputs a crisis instruction if the scratch area value is greater than the preset value N2; and outputs a correction instruction if the scratch area value is less than or equal to the preset value N2.

[0015] A rail train wheel surface scratch detection system, comprising: The wheel surface image acquisition module acquires the image of the wheel surface under high-speed motion captured by the visual camera under the illumination of the stroboscope as the wheel surface image; The abnormal feature mapping module detects the wheel surface images and the template images at different angles to obtain an abnormal feature group, then locates and splices the abnormal feature group to obtain an abnormal feature block, and maps the abnormal feature block to the front image of the wheel surface; A feature depth analysis module, which obtains depth information of a plurality of points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, wherein the plurality of points include edge points and non-edge points of the abnormal feature block, and compares the depth information of the edge points and the non-edge points; The abnormal screening and matching module outputs the abnormal type as a scratch type when the depth of the edge item point is less than the depth of the non-edge item point; otherwise, the abnormal type is output as other types.

[0016] The beneficial effects of the present invention are as follows: by combining the stroboscope with vision, a clear image can be captured when the train wheel surface is rotating. By utilizing change detection technology and template image comparison, abnormal features in the wheel surface image can be automatically identified and extracted, and then the abnormal features at different angles are spliced ​​to obtain a complete abnormal feature block that can be proved to be mapped in the front image of the wheel surface. Compared with the existing front wheel surface image, the splicing of different angles of the present invention can achieve the purpose of mutual verification, and it is difficult to completely wrap the wheel surface with a front image, which easily leads to feature detection missing; in addition, through the depth analysis of any item point on the edge contour of the abnormal feature block and the item point in the block area, it can be judged that the medium-low side height situation is the scratch target type, so as to meet the real-time monitoring of the wheel surface under high-speed driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is the overall flow chart of the present invention; Figure 2 is a flow chart of the feature verification analysis steps in the present invention; Figure 3 It is a flow chart of the instruction feedback step in the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The same parts are represented by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "bottom surface" and "top surface", "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0019] Since the current detection method cannot determine the abnormal defects on the wheel surface during movement, when scratches or stains appear on the wheel surface, thermal imaging cannot identify them, and ordinary vision is also difficult to determine, which is difficult to meet the requirements of surface scratch detection of the wheel surface. Therefore, the present invention designs a method for detecting scratches on the wheel surface of a railway train, such as Figure 1 As shown, the following steps are included: At least two sets of stroboscopes and matching visual cameras are installed on one side of the wheel surface of the bottom wheel of the train; The stroboscopic adjustment step is to index the corresponding stroboscope flashing frequency in a preset table as the current frequency value according to the train speed, wherein the train speed can be fed back by the train driving system, or the wheel speed can be calculated by detecting the wheel surface; The wheel surface image acquisition step is to obtain the image of the wheel surface under high-speed motion captured by the visual camera under the illumination of the stroboscope at the corresponding wheel speed as the wheel surface image. At this time, the single wheel surface image is a side axial view of the wheel surface. In addition, the wheel surface image needs to be preprocessed, including denoising, contrast enhancement, distortion correction, etc., to improve the accuracy of subsequent processing.

[0020] Abnormal feature mapping step, template image establishment: establish a standard template image of the wheel surface, the image should include a wheel surface without abnormal features, wherein the standard template image includes a side axial view template image of the wheel surface; Feature extraction: The wheel surface image at two basic angles and the corresponding template image are used for change detection to obtain the abnormal part. The main process of change detection is through image segmentation, feature extraction, change detection algorithm calculation and result verification. The existing change detection algorithms include direct comparison method, classification comparison method, difference method, ratio method and change vector analysis method. Abnormal feature positioning and stitching: stitch the abnormal feature parts located in the two wheel surface images to obtain a complete abnormal feature block, where the abnormal feature block is an abnormal block that can be mapped in the wheel surface front image, that is, the wheel surface abnormality map is obtained after mapping; The stitching principle is: edge detection, using edge detection algorithms (such as Canny) to extract the edge contour of each abnormal feature; feature matching, using matching algorithms (such as the nearest neighbor algorithm, k-nearest neighbor algorithm, etc.) to match the corresponding feature points between the two images to obtain a matching point group; calculating transformation, according to the matched feature points, calculating the transformation matrix between the images (such as affine transformation, perspective transformation, etc.), this transformation matrix describes how to map points on one image to corresponding points on another image; image stitching, using the calculated transformation matrix to stitch the two images, during the stitching process, using image fusion algorithms (such as linear fusion, pixel fusion, etc.) to process the gaps at the stitching to ensure that the transition area between the images is smooth, which is equivalent to aligning one of the abnormal features with another abnormal feature through a transformation matrix transformation, and after alignment, the two are fused to obtain an abnormal feature block; The mapping principle is as follows: calculate the transformation matrix between any wheel surface image and the front image of the wheel surface, select several edge contour points in the wheel surface image and transform them through the transformation matrix to obtain the position points of the edge contour points in the front image of the wheel surface, and then match the corresponding points of the abnormal feature block used for mapping through the position points, so that the abnormal feature block is mapped to the front image of the wheel surface as a whole. In the wheel surface image, find the vertices of the wheel surface as reference points and find the edge points of the abnormal features. Since the wheel surface image is a visually stretched image, the reference points and edge points can be determined. The reference points and edge points form a point group for calculating the transformation matrix, and then the corresponding position points in the abnormal feature block are mapped one by one to the front image of the wheel surface.

[0021] The feature depth analysis step obtains the depth information of several points in the abnormal feature block through the depth analysis strategy according to the abnormal feature block. The several points include edge points and non-edge points of the abnormal feature block. The depth information of the edge points and the non-edge points are compared. The depth analysis strategy includes an edge trajectory planning step and a depth estimation step. Edge trajectory planning step: since the abnormal feature block has been mapped in the front image of the wheel surface, several edge points of the abnormal feature block are drawn in the front image of the wheel surface to obtain an edge trajectory line, wherein the edge trajectory line is generated by using the drawn edge points to generate a continuous edge trajectory line through interpolation or fitting algorithm, and this edge trajectory line is used in the subsequent depth estimation step; In the depth estimation step, any point in the edge trajectory is selected as an edge point, and any point in the area surrounded by the edge trajectory is selected as a non-edge point. The parallax in the wheel surface image at any two different angles is calculated, and then the point distance between the item to be measured and the visual camera is obtained by depth calculation based on the parallax. The depth value of the item to be measured is obtained by subtracting the point distance from the distance of the visual camera from the wheel surface. Under normal circumstances, multiple edge points and non-edge points will be selected for calculation.

[0022] In the abnormal screening and matching step, when the depth of the edge item point is less than the depth of the non-edge item point, the edge is high and the middle is low, then the output abnormal type is the scratch type. On the contrary, the edge is low and the middle is high, then the output abnormal type is other types, including mud stain coverage type, etc.

[0023] When the train is running in the mountainous area, it rains or mud is crushed by the wheels, and the mud stains or adhere to the wheel surface. If it is a stone, after the stone is hit by the wheel, the stone will bounce and hit the wheel surface, causing scratches. Since the scratch itself is a concave area, when the mud stains cover the scratches, the wheels rotate at high speed, and the mud stains will spread, resulting in high edges and low middles. Therefore, it also includes feature verification analysis steps, such as Figure 2As shown, the wheel surface abnormality map is binarized to obtain a binary image. In the binary image, it is determined whether the pixel value in the abnormal feature block is the same as the wheel surface pixel value outside the abnormal feature block according to the pixel. If so, the output abnormality type is the scratch type. If not, the output abnormality type is the stain diffusion type. Specifically, if the pixel value in the abnormal feature block is the same as the pixel value of the normal part of the wheel surface (or very close, considering possible noise or error), it may indicate that the abnormality is caused by scratches or similar physical damages, because these damages usually destroy the original pixel value distribution, but may appear as the same value after binarization; If the pixel values ​​within the abnormal feature block are different from those of the normal part of the wheel surface, it may indicate that the abnormality is caused by stain diffusion, oil stains, watermarks, etc., because these types of abnormalities usually change the pixel values ​​so that they present different values ​​from the normal part of the wheel surface after binarization.

[0024] like Figure 2 As shown, it also includes a feature verification analysis step. When the pixel value in the abnormal feature block of the binary image is different from the wheel surface pixel value outside the abnormal feature block, the contour line of the abnormal feature block is obtained. If there is a sub-contour line in the contour area of ​​the abnormal feature block, the scratch filling type is output. The scratch filling type reflects the filling and diffusion of stains after the wheel surface is scratched. Specifically, sub-contour line detection is performed in the extracted contour line area of ​​the abnormal feature block. The sub-contour line refers to the additional contour line formed inside the abnormal feature block due to the filling of stains after the scratch. These sub-contour lines usually appear as small or irregular boundaries inside the abnormal feature block. If the sub-contour line is detected in the contour area of ​​the abnormal feature block, the scratch filling type is output. The scratch filling type reflects the situation that the wheel surface is filled and diffused by stains or other substances after the scratch. If the sub-contour line is not detected, it may belong to other types of abnormalities, such as simple scratches or stain diffusion, etc., and is output according to the previous analysis steps.

[0025] Since the wheel scratches are too serious, the bearing capacity of the wheel will be reduced, which may affect the driving safety. Therefore, the present invention also includes a command feedback step, such as Figure 3 As shown, the depth value is compared with the preset value N1. If the depth value is greater than the preset value N1, a crisis instruction is output to immediately notify relevant personnel or systems to take necessary measures, such as stopping for inspection, replacing the wheel surface, or performing emergency repairs, etc., to prevent potential safety accidents or performance degradation. If the depth value is less than or equal to the preset value N1, a correction instruction is output, which is only a recording function, and the relevant personnel or system will conduct targeted inspections and repairs on the wheel surface to ensure that it returns to normal.

[0026] like Figure 3As shown, the method also includes a scratch area calculation step, obtaining all abnormal feature blocks under the scratch type. A wheel surface on one side of a vehicle may include multiple scratch areas (abnormal feature blocks). Since the scratches are irregular, it is difficult to calculate the area of ​​the scratches. Therefore, the abnormal feature blocks are contour segmented to obtain several calculable scratch areas, and the area values ​​of the scratch areas are obtained by area calculation respectively, and the scratch area values ​​are accumulated to obtain the scratch area values. In the instruction feedback step, the scratch area value is compared with the preset value N2. If the scratch area value is greater than the preset value N2, a crisis instruction is output; if the scratch area value is less than or equal to the preset value N2, a correction instruction is output.

[0027] A rail train wheel surface scratch detection system corresponding to the scratch detection method includes: The wheel surface image acquisition module acquires the image of the wheel surface under high-speed motion captured by the visual camera under the illumination of the stroboscope as the wheel surface image; The abnormal feature mapping module detects changes in the wheel surface images and the template images at different angles to obtain abnormal feature groups, then locates and splices the abnormal feature groups to obtain abnormal feature blocks, and maps the abnormal feature blocks to the front image of the wheel surface; A feature depth analysis module, which obtains depth information of several points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, wherein the several points include edge points and non-edge points of the abnormal feature block, and compares the depth information of the edge points and the non-edge points; The abnormal screening and matching module outputs the abnormal type as the scratch type when the depth of the edge item point is less than the depth of the non-edge item point. Otherwise, the abnormal type is other types.

[0028] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for detecting scratches on a railway train wheel surface, characterized in that: The steps include: A wheel surface image acquisition step, in which a visual camera under the illumination of a stroboscope collects an image of the wheel surface under high-speed motion as the wheel surface image; Abnormal feature mapping step, obtaining an abnormal feature group by performing change detection on the wheel surface image and the template image at different angles, then locating and splicing the abnormal feature group to obtain an abnormal feature block, and mapping the abnormal feature block in the wheel surface front image as a wheel surface abnormality map; A feature depth analysis step, obtaining depth information of a plurality of points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, wherein the plurality of points include edge points and non-edge points of the abnormal feature block, and comparing the depth information of the edge points and the non-edge points; In the abnormal screening and matching step, when the depth of the edge item point is less than the depth of the non-edge item point, the output abnormal type is the scratch type, otherwise, the output abnormal type is other types.

2. A method for detecting scratches on a railway train wheel surface according to claim 1, characterized in that: The depth analysis strategy includes edge trajectory planning step and depth estimation step. The edge trajectory planning step is to draw a plurality of edge points of the abnormal feature block in the front image of the wheel surface to obtain an edge trajectory line; The depth estimation step selects any point in the edge trajectory as an edge point, selects any point in the area surrounded by the edge trajectory as a non-edge point, calculates the disparity in the wheel surface images at any two different angles, and then obtains the point distance between the item to be measured and the visual camera through depth calculation based on the disparity, and subtracts the point distance from the distance of the visual camera from the wheel surface to obtain the depth value of the item to be measured.

3. A method for detecting scratches on a railway train wheel surface according to claim 2, characterized in that: The abnormal feature mapping step extracts the edge contour of the abnormal feature according to edge detection, performs point matching on the edge contour of each abnormal feature in the abnormal feature group to obtain a matching point group, and then calculates a transformation matrix according to the matching point group, aligns one of the abnormal features with another abnormal feature through the transformation matrix, and after alignment, the two are fused to obtain an abnormal feature block.

4. A method for detecting scratches on a railway train wheel surface according to claim 3, characterized in that: The abnormal feature mapping step calculates a transformation matrix between any wheel surface image and the wheel surface front image, selects a number of edge contour points in the wheel surface image and transforms them through the transformation matrix to obtain the position points of the edge contour points in the wheel surface front image, and then matches the corresponding points used for mapping of the abnormal feature block through the position points, so that the abnormal feature block is mapped as a whole to the wheel surface front image.

5. A method for detecting scratches on a railway train wheel surface according to claim 4, characterized in that: The method also includes a stroboscopic adjustment step, in which the corresponding stroboscope flashing frequency is indexed in a preset table according to the train running speed as the current frequency value.

6. A method for detecting scratches on a railway train wheel surface according to claim 1 or 5, characterized in that: The method also includes a feature verification and analysis step, in which the wheel surface abnormality image is binarized to obtain a binary image, and in the binary image, it is judged based on pixels whether the pixel value in the abnormal feature block is the same as the wheel surface pixel value outside the abnormal feature block. If so, the output abnormality type is a scratch type, and if not, the output abnormality type is a stain diffusion type.

7. A rail vehicle wheel surface scratch detection method according to claim 6, characterized in that: It also includes a feature verification and analysis step. When the pixel value within the abnormal feature block in the binary image is different from the wheel surface pixel value outside the abnormal feature block, the contour line of the abnormal feature block is obtained. If there is a sub-contour line in the contour area of ​​the abnormal feature block, the scratch filling type is output, and the scratch filling type reflects the diffusion of stain filling after the wheel surface is scratched.

8. A method for detecting scratches on a railway train wheel surface according to claim 7, characterized in that: The method further includes an instruction feedback step, wherein the depth value is compared with a preset value N1, and if the depth value is greater than the preset value N1, a crisis instruction is output; if the depth value is less than or equal to the preset value N1, a correction instruction is output.

9. A method for detecting scratches on a railway train wheel surface according to claim 8, characterized in that: The method also includes a scratch area calculation step, obtaining an abnormal feature block under the scratch type, performing contour segmentation on the abnormal feature block to obtain a number of calculable scratch areas, respectively obtaining area values ​​of the scratch areas through area calculation, and accumulating them to obtain the scratch area value; The instruction feedback step compares the scratch area value with a preset value N2, and outputs a crisis instruction if the scratch area value is greater than the preset value N2; and outputs a correction instruction if the scratch area value is less than or equal to the preset value N2.

10. A railway train wheel surface scratch detection system, characterized in that: include The wheel surface image acquisition module acquires the image of the wheel surface under high-speed motion captured by the visual camera under the illumination of the stroboscope as the wheel surface image; The abnormal feature mapping module detects the wheel surface images and the template images at different angles to obtain an abnormal feature group, then locates and splices the abnormal feature group to obtain an abnormal feature block, and maps the abnormal feature block to the front image of the wheel surface; A feature depth analysis module, which obtains depth information of a plurality of points in the abnormal feature block through a depth analysis strategy according to the abnormal feature block, wherein the plurality of points include edge points and non-edge points of the abnormal feature block, and compares the depth information of the edge points and the non-edge points; The abnormal screening and matching module outputs the abnormal type as a scratch type when the depth of the edge item point is less than the depth of the non-edge item point; otherwise, the abnormal type is output as other types.

Images