A fastener pad model identification system and method

Abstract

This invention discloses a fastener pad model identification system and method, relating to the field of fastener maintenance technology. It includes an image acquisition module for acquiring fastener depth images from both sides of the rail; an image calibration module for height correction of the fastener depth images acquired from both sides of the sleeper where the rail is located; an image analysis module for preprocessing the corrected fastener depth images and segmenting the fastener components; a feature calculation module for extracting key region pixel values ​​and fitting depth parameters based on the corrected fastener depth images and segmented fastener component images to derive the fastener pad thickness; and a model identification module for comparing the fastener pad thickness with corresponding standard dimensions in a preset database to complete the fastener pad model identification. This invention uses machine vision inspection to detect the fastener pad thickness, thereby completing the fastener pad model identification, offering advantages of high detection efficiency and high identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of fastener maintenance technology, and in particular to a fastener pad model identification system and method. Background Technology

[0002] With the rapid development of rail transit technology in my country, it plays an increasingly important role in people's daily lives. Safety and comfort have always been topics of widespread public concern in the design and operation of rail transit. Therefore, strict requirements are placed on track smoothness.

[0003] In rail transit systems, when inspecting and adjusting rail smoothness, the height of the rail surface is typically adjusted by inserting shims of varying thicknesses at different sleeper locations on the rail base, based on the deformation data of the rail surface, to ensure rail smoothness. This requires knowing the current condition of the shims before adjustment. Due to the long time involved and multiple overlapping operations, the shim data usually needs to be rechecked to ensure the efficiency of the adjustment and maintenance work.

[0004] Currently, the work of verifying the pads is almost entirely done manually. During manual surveys, each surveyor needs to bend down close to the rail and use a ruler to read the pad thickness or visually identify the pad model for each sleeper to determine the pad thickness. This is labor-intensive, inefficient, and prone to significant reading errors. Summary of the Invention

[0005] In view of this, this application provides a fastener pad model identification system and method to address the shortcomings of the prior art.

[0006] The first aspect of this application provides a fastener pad model identification system, comprising:

[0007] The image acquisition module acquires images of the fastener depth on both sides of the rail;

[0008] The image calibration module performs height correction on the acquired fastener depth image based on the height of the sleeper support platform, and outputs the corrected fastener depth image.

[0009] The image analysis module preprocesses the corrected fastener depth image and segments the fastener components to obtain a component segmentation image;

[0010] The feature calculation module extracts pixel values ​​of key areas and fits and calculates depth parameters based on the corrected fastener depth image and the segmented image of the fastener assembly, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad.

[0011] The model identification module compares the derived fastener pad thickness with the corresponding standard size in the preset database to complete the fastener pad model identification.

[0012] In one possible implementation of the first aspect, extracting the pixel values ​​of the key region and fitting and calculating the depth parameters includes:

[0013] Based on the segmented image of the fastener assembly, the rail region, elastic bar region, and insulating block region are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the rail bottom.

[0014] The effective area of ​​the rail base after precise positioning is restored to the corrected fastener image, and two sets of pixel values ​​of the preset width area on both sides of the rail base are extracted. The preset width area is set based on the rail type.

[0015] The top surface of the rail base is obtained by linearly fitting the two sets of pixel values ​​using the least squares method. The top surface of the rail base is composed of inclined surfaces, the slope of which is determined by the rail type. The inclined surface area at a preset distance from the edge of the rail base is selected as the effective inclined surface.

[0016] Project the effective inclined plane onto the XOZ coordinate system and obtain the equation of the inclined line of the cross section; obtain the abscissa of the bottom edge of the rail and substitute it into the equation of the inclined line to calculate the depth coordinates of the two sets of inclined planes corresponding to the bottom of the rail.

[0017] The average depth coordinates of the corresponding inclined surfaces at the bottom of the two sets of rails are used to obtain the depth of the top surface of the rail bottom. ;

[0018] Based on the rail type, the preset rail base thickness is obtained, and the difference between the top surface depth of the rail base and the preset rail base thickness is calculated to obtain the bottom surface depth of the rail base. .

[0019] In one possible implementation of the first aspect, extracting the pixel values ​​of the key region and fitting and calculating the depth parameters further includes:

[0020] Based on the segmented image of the fastener assembly, the sleeper bearing platform area and the gauge baffle area are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the sleeper bearing platform.

[0021] The effective area of ​​the sleeper bearing platform after precise positioning is restored to the corrected fastener image, and the pixel values ​​of the sleeper bearing platform surface are extracted for plane fitting to obtain the bearing platform plane.

[0022] Project the rail support platform plane onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the first linear equation; obtain the abscissa of the rail support platform and substitute it into the first linear equation to calculate the depth of the rail support platform plane. .

[0023] In one possible implementation of the first aspect, extracting the pixel values ​​of the key region and fitting and calculating the depth parameters further includes:

[0024] Based on the segmented image of the fastener assembly, select the iron pad area and the spring strip area;

[0025] Based on the elastic bar region, an iron pad ROI region is generated in the iron pad region, and the iron pad ROI region is subjected to difference processing and morphological processing in sequence.

[0026] The ROI region of the iron pad after difference processing and morphological processing is restored to the corrected fastener image and grayscale morphological processing is performed to remove the pit area in the ROI region of the iron pad to obtain the effective area of ​​the iron pad.

[0027] The effective area of ​​the iron plate is located and restored to the corrected fastener image, and the pixel values ​​of the iron plate surface are extracted for plane fitting to obtain the iron plate plane.

[0028] Project the plane of the iron pad onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the second linear equation; obtain the abscissa of the iron pad and substitute it into the second linear equation to calculate the depth of the top surface of the iron pad. ;

[0029] Based on the type of iron pad, the preset thickness of the iron pad is obtained, and the difference between the top surface depth of the iron pad and the preset thickness is calculated to obtain the bottom surface depth of the iron pad. .

[0030] In one possible implementation of the first aspect, the thickness of the fastener pad is further derived to include:

[0031] The total thickness of the fastener pad is calculated using the following formula:

[0032]

[0033] This refers to the total thickness of the fastener pad.

[0034] The thickness of the shim on the fastener pad is calculated using the following formula:

[0035]

[0036] The thickness of the pad on the fastener pad;

[0037] The thickness of the lower shim plate of the fastener shim plate is calculated using the following formula:

[0038]

[0039] The thickness of the underside pad of the fastener pad.

[0040] In one possible implementation of the first aspect, identifying the fastener pad model includes:

[0041] The total thickness of the fastener pad is calculated. Thickness of the pad on the fastener pad and the thickness of the fastener pad plate under the pad plate The model of the fastener pad is identified by comparing it with the corresponding standard size in the preset database.

[0042] In one possible implementation of the first aspect, the image acquisition module specifically comprises:

[0043] The image acquisition module consists of one line structured light 3D camera, positioned directly above the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, two line structured light 3D cameras are used.

[0044] Alternatively, the image acquisition module may consist of two line structured light 3D cameras, symmetrically arranged above the fastener areas on both sides of the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, four line structured light 3D cameras may be used.

[0045] One possible implementation of the first aspect also includes:

[0046] The line structured light 3D camera includes a line structured light component and a 3D camera. The line structured light measurement algorithm built into the 3D camera is used to obtain the depth image of the fastener.

[0047] The depth measurement accuracy of the line structured light 3D camera is no less than 0.1 mm, and the horizontal measurement accuracy is no less than 0.1 mm / pixel.

[0048] In one possible implementation of the first aspect, preprocessing the corrected fastener depth image and segmenting the fastener assembly includes:

[0049] The corrected fastener image is first subjected to linear grayscale stretching and then standardization. A segmentation model is then used to segment the fastener components in the preprocessed fastener image.

[0050] One possible implementation of the first aspect also includes:

[0051] The linear grayscale stretching process is performed using the aforementioned formula, specifically as follows:

[0052]

[0053] This is a diagram showing the result after linear stretching. This is the corrected fastener depth diagram. The maximum value in the corrected fastener depth diagram. To obtain the minimum value in the corrected fastener depth diagram, This represents the maximum value in the fastener drawing after grayscale stretching. This is the minimum value in the fastener drawing after grayscale stretching;

[0054] The standardization process is performed using the aforementioned standardization formula, specifically as follows:

[0055]

[0056] in, The input data before standardization. For standardized data, The mean of all image data for ballast track is given. The standard deviation of all image data for ballast track;

[0057] The segmentation model used is the SINet neural network segmentation model.

[0058] A second aspect of this application provides a method for identifying the model of a fastener pad, including:

[0059] Collect images of the fastener depth on both sides of the rail;

[0060] Based on the height of the sleeper support platform, the acquired fastener depth image is height-corrected, and the corrected fastener depth image is output.

[0061] The corrected fastener depth image is preprocessed and the fastener components are segmented to obtain the component segmentation image;

[0062] Based on the corrected fastener depth image and the segmented image of the fastener assembly, the pixel values ​​of key areas are extracted and depth parameters are fitted and calculated, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad.

[0063] The derived fastener pad thickness is compared with the corresponding standard size in the preset database to complete the fastener pad model identification.

[0064] Its beneficial effects are as follows: This invention discloses a fastener pad model recognition system and method, comprising an image acquisition module that uses fastener depth images from both sides of the rail; an image stitching and calibration module for height correction of the fastener depth images acquired from both sides of the sleeper where the rail is located; an image analysis module for preprocessing the corrected fastener depth images and segmenting the fastener components; a feature calculation module for extracting key region pixel values ​​and fitting and calculating depth parameters based on the corrected fastener depth images and the segmented fastener component images, thereby deriving the fastener pad thickness; and a model recognition module for comparing the fastener pad thickness with the corresponding standard size in a preset database to complete the fastener pad model recognition. This invention uses machine vision inspection to detect the fastener pad thickness, thereby completing the fastener pad model recognition, which has the advantages of high detection efficiency and high recognition accuracy. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of a fastener pad model identification system provided in an embodiment of this application;

[0067] Figure 2 This is a schematic diagram of the image acquisition module in an embodiment of this application;

[0068] Figure 3 This is a schematic diagram of the image stitching and calibration module in an embodiment of this application;

[0069] Figure 4 This is a flowchart illustrating the reasoning process of the image analysis module in an embodiment of this application.

[0070] Figure 5 This is a schematic diagram illustrating the calculation of fastener pad thickness provided in an embodiment of this application;

[0071] Figure 6 This is a schematic flowchart of a fastener pad model identification method provided in an embodiment of this application. Detailed Implementation

[0072] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0073] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0074] Example 1

[0075] In rail transit systems, when inspecting and adjusting rail smoothness, the height of the rail surface is typically adjusted by inserting shims of varying thicknesses at different sleeper locations on the rail base, based on the deformation data of the rail surface, to ensure rail smoothness. This requires knowing the current status data of the shims before adjustment. Due to the long time involved and multiple overlapping operations, the shim data usually needs to be rechecked to ensure the efficiency of the adjustment and maintenance work. Currently, this part of the shim verification work is almost entirely done manually. During manual surveying, each surveyor needs to bend down close to the rail and use a ruler to read the shim thickness of each sleeper's shim or visually identify the shim type to determine the thickness. This is labor-intensive, inefficient, and prone to significant reading errors.

[0076] Therefore, this application provides a fastener pad model identification system, such as... Figure 1 As shown, it includes:

[0077] The image acquisition module acquires images of the fastener depth on both sides of the rail;

[0078] The image calibration module performs height correction on the acquired fastener depth image based on the height of the sleeper support platform, and outputs the corrected fastener depth image.

[0079] The image analysis module preprocesses the corrected fastener depth image and segments the fastener components to obtain a component segmentation image;

[0080] The feature calculation module extracts pixel values ​​of key areas and fits and calculates depth parameters based on the corrected fastener depth image and the segmented image of the fastener assembly, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad.

[0081] The model identification module compares the derived fastener pad thickness with the corresponding standard size in the preset database to complete the fastener pad model identification.

[0082] Image acquisition module (such as) Figure 2 (as shown)

[0083] The image acquisition module consists of one line structured light 3D camera, positioned directly above the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, two line structured light 3D cameras are used.

[0084] Alternatively, the image acquisition module consists of two line structured light 3D cameras, symmetrically arranged above the fastener areas on both sides of the rail, to image the fastener areas on both sides of the rail. To achieve the recognition of the rail areas on both sides, four line structured light 3D cameras are used.

[0085] The line structured light 3D camera contains a line structured light component and a 3D camera. It uses the line structured light measurement algorithm built into the 3D camera to obtain the fastener depth image.

[0086] The depth measurement accuracy of the line structured light 3D camera is no less than 0.1mm, and the horizontal measurement accuracy is no less than 0.1mm / pixel.

[0087] Furthermore, the line structured light 3D camera can be replaced with a surface structured light 3D scanner to acquire fastener depth images; it can also be replaced with a 3D imaging camera based on binocular vision and laser speckle to acquire fastener depth images, with the camera layout being the same as that of the line structured light 3D camera.

[0088] Image calibration module (such as) Figure 3 (as shown)

[0089] When the images of the fasteners on both sides of the rail are captured by two line structured light 3D cameras, the two fastener images need to be stitched together. The two fastener images on both sides of the sleeper where the rail is located are stitched together according to the width of the rail base (a total of four fastener images for the left and right rails). This embodiment uses a 60-gauge rail as an example, with a standard rail base width of 150mm and an image acquisition resolution of 0.1mm / pixel in the x-direction. After edge detection of the fastener images, the rail edges of the two fastener images are extracted, and the distance between the rail edges (e.g., ...) is calculated. Figure 3As shown, the distance between the two straight lines a and b should be controlled to be equal to the width of the rail base (150mm). In other words, the distance should be controlled within 1500 pixels to complete the splicing of the two fastener images.

[0090] Slight camera tilt during shooting and minor errors in the sleeper's installation height can cause misalignment in the y-direction (height direction) between the left and right fastener images. For example, the rail support platform in one image might be higher or lower than in the other. Failure to correct this can lead to distortion in subsequent depth calculations. The height of the sleeper rail support platform is a fixed design parameter (physical standard value) in rail transit. Using this height as a benchmark, by comparing the pixel positions of the sleeper rail support platform in the left and right fastener images and adjusting the y-direction offset of one image, the height of the rail support platform in both images is made consistent. Essentially, this uses a physical standard to correct the vertical deviation of the images, ensuring that the stitched fastener image conforms to the actual working conditions in the height direction.

[0091] Image analysis module (such as) Figure 4 (as shown)

[0092] The linear grayscale stretching process is performed using the aforementioned formula, specifically as follows:

[0093]

[0094] This is a diagram showing the result after linear stretching. This is the corrected fastener depth diagram. The maximum value in the corrected fastener depth diagram. To obtain the minimum value in the corrected fastener depth diagram, This represents the maximum value in the fastener drawing after grayscale stretching. This is the minimum value in the fastener drawing after grayscale stretching;

[0095] The standardization process is performed using the aforementioned standardization formula, which is as follows:

[0096]

[0097] in, The input data before standardization. For standardized data, The mean of all image data for ballast track is given. The standard deviation of all image data for ballast track;

[0098] The preprocessed data is then input into the segmentation model. In this embodiment, the SINet neural network segmentation model is used. The model automatically completes the segmentation of the fastener components and then performs post-processing on the results to ensure accurate segmentation and prepare data for the next step.

[0099] Feature calculation module (e.g.) Figure 5 (as shown)

[0100] Based on the stitched and corrected fastener image and the fastener assembly image derived from the segmentation model, pixel values ​​of key regions are extracted and depth parameters are calculated. Based on all calculated depth parameters, the thickness of the fastener pad is derived. Key regions include the rail region, elastic clip region, insulating block region, sleeper support platform region, gauge baffle region, and iron pad region. Depth parameters include the depth of the rail base. Depth of the rail support platform Depth of top surface of iron pad Depth of the bottom surface of the iron pad .

[0101] 1. Regarding the depth of the bottom surface of the rail: Derivation

[0102] The core principle combines machine vision, geometric fitting, and industry-standard mapping. First, it accurately locates the effective area through image preprocessing, then restores the pixel data to the geometric shape of physical space, and finally uses the industry-standard parameters of the rail to derive the key depth, providing a precise benchmark for subsequent calculations of the base plate. The breakdown is as follows:

[0103] Region preprocessing utilizes difference processing and morphological processing to purify the effective regions. Although the segmentation model can identify the rail region, elastic bar region, and insulation block region, these three regions overlap (e.g., elastic bar occluding the rail) or have image noise (e.g., small spots). Differential processing and morphological processing are needed to remove invalid information. Difference processing essentially preserves the target and removes interference, accurately separating the unobstructed, pure rail base effective region from the rail region, elastic bar region, and insulation block region, avoiding pixel interference from elastic bar and insulation block pixels in subsequent calculations. Morphological processing is equivalent to image denoising and edge regularization, eliminating tiny noise points in the fastener component image, filling in regional gaps, making the rail base edge clearer, and ensuring that the subsequently extracted pixels are effective signals rather than interfering noise.

[0104] It should be noted that difference processing is essentially an image segmentation optimization technique that removes interfering regions from the target region, retaining only the pure, valid target region. Its core purpose is to eliminate invalid information caused by component occlusion and region overlap, providing accurate region localization for subsequent pixel extraction and depth calculation. The specific difference processing logic is as follows:

[0105] Based on the segmented image of the fastener assembly output by the segmentation model, a target region (such as the rail region, rail support platform region, or iron pad ROI region) is first selected.

[0106] Then select one or more interference areas (such as the spring bar area, the insulating block area, the track gauge baffle area).

[0107] The target region and interference region are calculated by image algorithm, and finally the pure target effective region is obtained by removing the interference region. This is equivalent to removing occlusion and impurities from the target region and retaining only the core part that needs to be detected.

[0108] Scenario 1: Purification of the effective area at the bottom of the rail

[0109] Target region: Rail region in the segmentation image; Interference region: Elastic bar region, insulating block region; Difference operation: Rail region - Elastic bar region - Insulating block region; Purpose: To remove the occlusion of the rail base by the elastic bar and insulating block, and retain the uncovered pure rail base region, avoiding pixel interference from occlusion objects in subsequent slope fitting and depth calculation.

[0110] Scenario 2: Purification of the effective area of ​​the sleeper support platform

[0111] Target region: The sleeper support platform region in the segmentation image; Interference region: The gauge baffle region; Difference operation: Support platform region - Gauge baffle region; Function: Eliminate the occlusion of the gauge baffle on the support platform surface, ensuring that the extracted pixels come only from the support platform itself, and guaranteeing the accuracy of plane fitting.

[0112] Scenario 3: Purification of the effective area of ​​the iron pad

[0113] Target region: The iron pad ROI region generated based on the spring bar region; Interference region: the spring bar region; Difference operation: Iron pad ROI region - spring bar region; Function: Remove the occluded part of the spring bar within the iron pad region, retain the iron pad body region, and lay the foundation for subsequent grayscale morphological processing.

[0114] Pixel extraction: During image acquisition, the correspondence between pixels and physical dimensions has been defined. In this embodiment, the top surface of the 60-track bottom is an industrial standard 1:9 slope (extending 9mm horizontally and decreasing 1mm in depth), with an effective width of 25mm (physical dimension). This corresponds to 250 pixels in the image, which is 25mm ÷ 0.1mm / pixel = 250 pixels. Extracting the pixel values ​​of a 25mm wide area on each side edge of the track bottom essentially locks down the effective pixel set that reflects the 1:9 slope shape (one set on each side, for a total of two sets).

[0115] The inclined plane fitting process involves fitting the top surface of the rail base, which is a standard 1:9 inclined plane in the industry. Therefore, the two sets of discrete pixels extracted will theoretically follow the geometric rules of this inclined plane. By using the least squares method, these discrete pixels are fitted into a continuous inclined line, which is the mathematical expression of the inclined plane. The fitted inclined plane is then projected into the XOZ coordinate system (ignoring the length direction y and only focusing on the cross-section), resulting in the equation of the inclined line (Z=kx+b, k=-1 / 9, corresponding to the 1:9 inclined plane, b is the intercept), which transforms the visual morphology in the image into a computable mathematical model.

[0116] It should be noted that the XOZ coordinate system (e.g.) Figure 5 The diagram shows a two-dimensional cross-sectional coordinate system used to convert image pixel information into physical space depth parameters. It focuses on two key dimensions: horizontal position and depth, providing a unified computational framework for plane / slope fitting and depth value derivation. The X-axis represents the horizontal direction, corresponding to the transverse direction of the stitched and corrected image (i.e., the width direction of the rail and the transverse direction of the sleeper); the Z-axis represents the depth direction, perpendicular to the height / depth of the track surface, directly corresponding to the depth value in physical space; the Y-axis represents the length direction, corresponding to the extension direction parallel to the rail. Since this embodiment only focuses on the depth relationship of the cross-section, the Y-axis is ignored during projection to reduce computational complexity.

[0117] Depth calculation involves substituting the x-coordinate of the rail base edge into the equation of the inclined plane to obtain the Z-coordinate (depth value) of the two inclined planes at the edge position; the average of these two values ​​is then taken to offset random errors caused by single-pixel noise and shooting deviation, thus determining the depth of the top surface of the rail base. This is closer to the true value; however, since rails are standardized industrial parts, their rail base thickness is a fixed value (e.g., the rail base thickness of a 60 rail is a fixed parameter). Because the bottom surface of the rail base may be obscured by pads or oil stains, pixel values ​​cannot be directly extracted. Therefore, an indirect derivation is used, using the depth of the top surface of the rail base. Subtract the fixed rail base thickness to obtain the rail base depth. .

[0118] It should be noted that the principle of obtaining the top surface depth of the rail bottom using the effective inclined plane is as follows: First, fit the effective inclined plane and then project it onto the XOZ coordinate system to obtain the equation of the effective inclined plane. Then, substitute the x-coordinates of the bottom edges of the rail bottom on both sides (the bottom edges of the effective inclined plane) into the equation of the inclined plane to obtain the Z-coordinates (i.e., depth values) of the inclined plane on both sides at the edge positions. Take the average of the two values ​​as the top surface depth value of the rail bottom.

[0119] 2. Regarding the depth of the track support platform Derivation

[0120] The core principle is region purification, pixel-to-physical space mapping, and planar fitting modeling. First, the effective area of ​​the support platform is locked through image preprocessing. Then, the image pixel data is restored to the planar shape of physical space. Finally, the precise depth of the top surface of the support platform is calculated through the data model, providing a reliable benchmark for the calculation of the pad thickness. The breakdown is as follows:

[0121] Region preprocessing involves extracting the effective area of ​​the rail bearing platform through difference processing and morphological processing. Difference processing precisely separates the pure rail bearing platform area unobstructed by the gauge baffle, utilizing the rail bearing platform area and the gauge baffle area, thus avoiding the influence of occluding pixels on subsequent depth calculations. Morphological processing converts the difference-processed area into a black-and-white binary image, filling small gaps through dilation and eliminating small noise points through erosion, making the edge contour of the rail bearing platform clearer and ensuring that the subsequently extracted pixels are valid signals from the rail bearing platform surface. This final effective area of ​​the rail bearing platform provides the foundation for extracting accurate pixel values.

[0122] Pixel extraction: The effective area of ​​the sleeper bearing platform in the fastener assembly image only provides position coordinates. The coordinate information of this area needs to be mapped to the corrected fastener image. Therefore, the stitched and corrected fastener image has established a precise correspondence between pixel coordinates and physical dimensions, and eliminated stitching misalignment and height error. From the effective area of ​​the stitched and corrected bearing platform, the depth coordinates Z and horizontal coordinates X of all pixels are extracted. Each pixel corresponds to a real point on the surface of the bearing platform in physical space. The set of these points constitutes a complete pixel sample of the bearing platform surface.

[0123] Plane fitting: The rail sleeper support platform is an industrial prefabricated component, and its top surface is designed as a standard plane. Therefore, the extracted discrete pixels will theoretically be distributed on the same plane. The least squares method is used to perform plane fitting on all discrete pixels to obtain the rail support platform plane. The fitted rail support platform plane is projected into the XOZ coordinate system to obtain the linear equation of the cross section, thus transforming the visual surface in the image into a computable mathematical model.

[0124] For depth calculation, the rail support platform is a continuous plane, and its depth value varies with the horizontal coordinate X according to the above-mentioned linear equation. Therefore, by obtaining any horizontal coordinate within the effective area of ​​the rail support platform and substituting it into the linear equation, the Z coordinate (depth value) of the corresponding position can be calculated, which is the plane depth of the rail support platform. .

[0125] 3. Regarding the depth of the top surface of the iron pad Depth of the bottom surface of the iron pad Derivation

[0126] The core principle involves precise ROI region locking, multi-level image purification, pixel-to-physical space mapping, planar fitting modeling, and indirect derivation based on industry standards. First, multi-step image processing locks the interference-free effective area of ​​the iron pad. Then, pixel data is restored to a planar shape in physical space. Finally, the bottom depth is derived using the industry-standard thickness of the iron pad, providing a precise benchmark for calculating the thickness of the upper and lower pads. The breakdown is as follows:

[0127] ROI generation: The iron pad and the spring strip are supporting components in the fastener system. The installation position of the spring strip and the iron pad have a fixed spatial relationship. Using the spring strip area identified in the fastener component image as the positioning anchor point, the ROI (Region of Interest) of the iron pad is generated. Essentially, it uses the fixed spatial relationship between the two to quickly delineate the approximate range of the iron pad, avoiding the positioning offset of the iron pad area caused by interference from other components in the fastener image.

[0128] Multi-level image purification utilizes difference processing, morphological processing, and grayscale morphological processing to eliminate interference. The process involves several steps: First, difference processing is used to remove the portion of the spring bar that is obscured by the iron pad's ROI and the spring bar region. This ensures that only the iron pad itself, not covered by the spring bar, is retained, preventing pixel interference with depth calculation. Second, morphological processing (in this embodiment, morphological processing refers to binary morphological processing) converts the difference-processed region into a black-and-white binary image. Dilation and erosion operations are then used to refine the edge contours of the iron pad, removing noise such as burrs and spots from the fastener assembly image, resulting in clearer region boundaries. Third, grayscale morphological processing locates the processed iron pad ROI in the stitched and corrected fastener image. Morphological operations are then performed on the image's grayscale values. Since pits on the iron pad surface are local defects (not signals related to actual thickness), this operation filters out pixels corresponding to these pits, retaining only the effective pixels on the flat surface of the iron pad. This prevents local defects from distorting depth calculations, ultimately yielding the effective area of ​​the iron pad and ensuring that subsequently extracted pixels reflect the true surface of the iron pad.

[0129] After pixel extraction and splicing correction, a precise correspondence between pixel coordinates and physical dimensions has been established in the fastener image, and splicing misalignment and height deviation have been eliminated. The coordinates of the effective area of ​​the iron pad are mapped to the spliced ​​and corrected fastener image, and the horizontal and depth coordinates of all pixels in the area are extracted. Each pixel corresponds to a real point on the surface of the iron pad in physical space. The set of these points constitutes a complete pixel sample of the surface of the iron pad.

[0130] Plane fitting: The iron pad is a standardized industrial prefabricated part, and its top surface is designed as a standard plane. Therefore, the extracted discrete pixels will theoretically follow the geometric laws of the plane. The least squares method is used to fit the plane to all discrete pixels, and the fitted plane is projected into the XOZ coordinate system to obtain the straight line equation of the cross section, thus transforming the visual surface in the image into a computable mathematical model.

[0131] For depth calculation, the top surface of the iron pad is a continuous plane, and its depth value varies with the horizontal coordinate according to the aforementioned linear equation. By obtaining any horizontal coordinate within the effective area of ​​the iron pad and substituting it into the linear equation, the Z coordinate of the corresponding position can be calculated, which is the depth of the top surface of the iron pad. Since planar fitting has offset individual pixel noise and minor surface imperfections, To achieve the average effective depth and reflect the true height, the iron pad is a standardized industrial part with a fixed thickness. Since the bottom surface of the iron pad is in contact with the fastener pad, it may be obscured by the fastener pad, oil, or impurities, making direct pixel value extraction impossible. Therefore, an indirect derivation is used, employing the depth of the top surface of the iron pad. Subtracting the fixed thickness of the iron pad, we obtain the depth of the bottom surface of the iron pad. .

[0132] 4. Derivation of the thickness of the fastener pad

[0133] It should be noted that the fastener pad is different from the iron pad. In terms of the physical spatial relationship of the track, it should be the sleeper support platform, the lower pad of the fastener pad, the iron pad, the lower pad of the fastener pad, and the rail. The upper pad of the fastener pad, the iron pad, and the lower pad of the fastener pad together constitute the fastener pad.

[0134] Therefore, the total thickness of the fastener pad is calculated using the following formula:

[0135]

[0136] This refers to the total thickness of the fastener pad.

[0137] The total thickness of the fastener pad is equal to the sum of the thickness of the pad on the upper part of the fastener pad, the thickness of the iron pad, and the thickness of the pad below the fastener pad. .

[0138] The thickness of the shim on the fastener pad is calculated using the following formula:

[0139]

[0140] The thickness of the pad on the fastener pad;

[0141] The thickness of the lower shim plate of the fastener shim plate is calculated using the following formula:

[0142]

[0143] The thickness of the underside pad of the fastener pad.

[0144] Model identification module:

[0145] The core principle is feature matching and comparison with a standard database. Essentially, it involves accurately matching the detected physical features with preset model-size mapping rules, eliminating interference through quantitative comparison, and finally determining the unique corresponding pad model. The core logic can be broken down as follows:

[0146] A mapping table of fastener pad models and corresponding standard thickness parameters is pre-stored, with core data sourced from the fastener pad industry standards in the rail transit sector;

[0147] Three core thickness parameters are obtained from the feature calculation module. These parameters are actual physical dimensions obtained based on image fitting and depth derivation, rather than abstract image pixel values, and have a basis for direct comparison with standard dimensions.

[0148] A single thickness parameter may result in different models having the same total thickness but different layer thicknesses (e.g., model B: , , Model C: , , ), only comparison This module may make incorrect judgments; it also compares... , and Three dimensions are used to ensure the uniqueness of the match through multi-dimensional constraints, thus solving the problem of misjudgment in single feature matching.

[0149] Essentially, it transforms model identification into a precise matching problem of quantified dimensions. By using a standardized database and multi-dimensional comparison rules, it replaces manual visual identification, ensuring both accuracy and significantly improving efficiency, thus solving pain points in manual identification such as easily confused models.

[0150] This embodiment overcomes the visual limitations of manual inspection by integrating machine vision and segmentation models; it achieves accurate derivation of the layer thickness of fastener pads through a multi-dimensional depth calculation model; it solves the efficiency and accuracy problems of manual inspection through standardized and automated process design; and it adapts to the trend of intelligent maintenance in rail transit through data closed-loop design. Compared with existing technologies, the technological advancement of this embodiment is not an optimization of a single link, but rather the construction of a complete chain of technologies including image acquisition, component recognition, depth calculation, model matching, and data traceability. It fundamentally solves the problems of low efficiency and poor accuracy in existing manual inspection, realizing the development from manual experience-based inspection to intelligent data-based inspection. It provides an efficient, accurate, and safe technical solution for rail transit fastener maintenance, and has significant engineering application value and industrialization prospects.

[0151] In some embodiments, extracting pixel values ​​of key regions and fitting and calculating depth parameters includes:

[0152] Based on the segmented image of the fastener assembly, the rail region, elastic bar region, and insulating block region are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the rail bottom.

[0153] The effective area of ​​the rail base after precise positioning is restored to the corrected fastener image, and two sets of pixel values ​​of the preset width area on both sides of the rail base are extracted. The preset width area is set based on the rail type.

[0154] The top surface of the rail base is obtained by linearly fitting the two sets of pixel values ​​using the least squares method. The top surface of the rail base is composed of inclined surfaces, the slope of which is determined by the rail type. The inclined surface area at a preset distance from the edge of the rail base is selected as the effective inclined surface.

[0155] Project the effective inclined plane onto the XOZ coordinate system and obtain the equation of the inclined line of the cross section; obtain the abscissa of the bottom edge of the rail and substitute it into the equation of the inclined line to calculate the depth coordinates of the two sets of inclined planes corresponding to the bottom of the rail.

[0156] The average depth coordinates of the corresponding inclined surfaces at the bottom of the two sets of rails are used to obtain the depth of the top surface of the rail bottom. ;

[0157] Based on the rail type, the preset rail base thickness is obtained, and the difference between the top surface depth of the rail base and the preset rail base thickness is calculated to obtain the bottom surface depth of the rail base. .

[0158] In some embodiments, extracting pixel values ​​of key regions and fitting and calculating depth parameters further includes:

[0159] Based on the segmented image of the fastener assembly, the sleeper bearing platform area and the gauge baffle area are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the sleeper bearing platform.

[0160] The effective area of ​​the sleeper bearing platform after precise positioning is restored to the corrected fastener image, and the pixel values ​​of the sleeper bearing platform surface are extracted for plane fitting to obtain the bearing platform plane.

[0161] Project the rail support platform plane onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the first linear equation; obtain the abscissa of the rail support platform and substitute it into the first linear equation to calculate the depth of the rail support platform plane. .

[0162] In some embodiments, extracting pixel values ​​of key regions and fitting and calculating depth parameters further includes:

[0163] Based on the segmented image of the fastener assembly, select the iron pad area and the spring strip area;

[0164] Based on the elastic bar region, an iron pad ROI region is generated in the iron pad region, and the iron pad ROI region is subjected to difference processing and morphological processing in sequence.

[0165] The ROI region of the iron pad after difference processing and morphological processing is restored to the corrected fastener image and grayscale morphological processing is performed to remove the pit area in the ROI region of the iron pad to obtain the effective area of ​​the iron pad.

[0166] The effective area of ​​the iron plate is located and restored to the corrected fastener image, and the pixel values ​​of the iron plate surface are extracted for plane fitting to obtain the iron plate plane.

[0167] Project the plane of the iron pad onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the second linear equation; obtain the abscissa of the iron pad and substitute it into the second linear equation to calculate the depth of the top surface of the iron pad. ;

[0168] Based on the type of iron pad, the preset thickness of the iron pad is obtained, and the difference between the top surface depth of the iron pad and the preset thickness is calculated to obtain the bottom surface depth of the iron pad. .

[0169] In some embodiments, the thickness of the fastener pad is further derived to include:

[0170] The total thickness of the fastener pad is calculated using the following formula:

[0171]

[0172] This refers to the total thickness of the fastener pad.

[0173] The thickness of the shim on the fastener pad is calculated using the following formula:

[0174]

[0175] The thickness of the pad on the fastener pad;

[0176] The thickness of the lower shim plate of the fastener shim plate is calculated using the following formula:

[0177]

[0178] The thickness of the underside pad of the fastener pad.

[0179] In some embodiments, identifying the fastener pad model includes:

[0180] The total thickness of the fastener pad is calculated. Thickness of the pad on the fastener pad and the thickness of the fastener pad plate under the pad plate The model of the fastener pad is identified by comparing it with the corresponding standard size in the preset database.

[0181] In some embodiments, the image acquisition module specifically comprises:

[0182] The image acquisition module consists of one line structured light 3D camera, positioned directly above the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, two line structured light 3D cameras are used.

[0183] Alternatively, the image acquisition module may consist of two line structured light 3D cameras, symmetrically arranged above the fastener areas on both sides of the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, four line structured light 3D cameras may be used.

[0184] In some embodiments, it also includes:

[0185] The line structured light 3D camera includes a line structured light component and a 3D camera. The line structured light measurement algorithm built into the 3D camera is used to obtain the depth image of the fastener.

[0186] The depth measurement accuracy of the line structured light 3D camera is no less than 0.1 mm, and the horizontal measurement accuracy is no less than 0.1 mm / pixel.

[0187] In some embodiments, preprocessing the corrected fastener depth image and segmenting the fastener assembly includes:

[0188] The corrected fastener image is first subjected to linear grayscale stretching and then standardization. A segmentation model is then used to segment the fastener components in the preprocessed fastener image.

[0189] In some embodiments, it also includes:

[0190] The linear grayscale stretching process is performed using the aforementioned formula, specifically as follows:

[0191]

[0192] This is a diagram showing the result after linear stretching. This is the corrected fastener depth diagram. The maximum value in the corrected fastener depth diagram. To obtain the minimum value in the corrected fastener depth diagram, This represents the maximum value in the fastener drawing after grayscale stretching. This is the minimum value in the fastener drawing after grayscale stretching;

[0193] The standardization process is performed using the aforementioned standardization formula, specifically as follows:

[0194]

[0195] in, The input data before standardization. For standardized data, The mean of all image data for ballast track is given. The standard deviation of all image data for ballast track;

[0196] The segmentation model used is the SINet neural network segmentation model.

[0197] Example 2

[0198] Based on the fastener pad model identification system provided in Embodiment 1 of this application, correspondingly, Embodiment 2 of this application also provides a fastener pad model identification method, such as... Figure 6 As shown, it includes:

[0199] Collect images of the fastener depth on both sides of the rail;

[0200] Based on the height of the sleeper support platform, the acquired fastener depth image is height-corrected, and the corrected fastener depth image is output.

[0201] The corrected fastener depth image is preprocessed and the fastener components are segmented to obtain the component segmentation image;

[0202] Based on the corrected fastener depth image and the segmented image of the fastener assembly, the pixel values ​​of key areas are extracted and depth parameters are fitted and calculated, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad.

[0203] The derived fastener pad thickness is compared with the corresponding standard size in the preset database to complete the fastener pad model identification.

[0204] In the process of acquiring images of rail fasteners, a light source and supplementary lighting are used to illuminate the fasteners to be photographed. Two line structured light 3D cameras capture two fastener images from each side of the rail, providing a data foundation for subsequent identification of the fastener pad type on each side of the rail. Furthermore, using four line structured light 3D cameras, four fastener images can be obtained from both sides of the rail, providing a data foundation for simultaneous identification of the fastener pad type on both sides of the rail.

[0205] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0206] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0207] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A fastener pad model identification system, characterized in that, include: The image acquisition module acquires images of the fastener depth on both sides of the rail; The image calibration module performs height correction on the acquired fastener depth image based on the height of the sleeper support platform, and outputs the corrected fastener depth image. The image analysis module preprocesses the corrected fastener depth image and segments the fastener assembly to obtain a segmented image of the fastener assembly. The feature calculation module extracts pixel values ​​of key areas and fits and calculates depth parameters based on the corrected fastener depth image and the segmented image of the fastener assembly, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad. The model identification module compares the derived fastener pad thickness with the corresponding standard size in the preset database to complete the fastener pad model identification. Extracting pixel values ​​from key regions and fitting and calculating depth parameters includes: Based on the segmented image of the fastener assembly, the rail region, elastic bar region, and insulating block region are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the rail bottom. The effective area of ​​the rail base after precise positioning is restored to the corrected fastener image, and two sets of pixel values ​​of the preset width area on both sides of the rail base are extracted. The preset width area is set based on the rail type. The top surface of the rail base is obtained by linearly fitting the two sets of pixel values ​​using the least squares method. The top surface of the rail base is composed of inclined surfaces, the slope of which is determined by the rail type. The inclined surface area at a preset distance from the edge of the rail base is selected as the effective inclined surface. Project the effective inclined plane onto the XOZ coordinate system and obtain the equation of the inclined line of the cross section; obtain the abscissa of the bottom edge of the rail and substitute it into the equation of the inclined line to calculate the depth coordinates of the two sets of inclined planes corresponding to the bottom of the rail. The average depth coordinates of the corresponding inclined surfaces at the bottom of the two sets of rails are used to obtain the depth of the top surface of the rail bottom. ; Based on the rail type, the preset rail base thickness is obtained, and the difference between the top surface depth of the rail base and the preset rail base thickness is calculated to obtain the bottom surface depth of the rail base. .

2. The fastener pad model identification system according to claim 1, characterized in that, Extracting pixel values ​​from key regions and fitting and calculating depth parameters also includes: Based on the segmented image of the fastener assembly, the sleeper bearing platform area and the gauge baffle area are selected and subjected to difference processing and morphological processing to achieve accurate positioning of the effective area of ​​the sleeper bearing platform. The effective area of ​​the sleeper bearing platform after precise positioning is restored to the corrected fastener image, and the pixel values ​​of the sleeper bearing platform surface are extracted for plane fitting to obtain the bearing platform plane. Project the rail support platform plane onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the first linear equation; obtain the abscissa of the rail support platform and substitute it into the first linear equation to calculate the depth of the rail support platform plane. .

3. The fastener pad model identification system according to claim 2, characterized in that, Extracting pixel values ​​from key regions and fitting and calculating depth parameters also includes: Based on the segmented image of the fastener assembly, select the iron pad area and the spring strip area; Based on the elastic bar region, an iron pad ROI region is generated in the iron pad region, and the iron pad ROI region is subjected to difference processing and morphological processing in sequence. The ROI region of the iron pad after difference processing and morphological processing is restored to the corrected fastener image and grayscale morphological processing is performed to remove the pit area in the ROI region of the iron pad to obtain the effective area of ​​the iron pad. The effective area of ​​the iron plate is located and restored to the corrected fastener image, and the pixel values ​​of the iron plate surface are extracted for plane fitting to obtain the iron plate plane. Project the plane of the iron pad onto the XOZ coordinate system and obtain the linear equation of the cross-section, denoted as the second linear equation; obtain the abscissa of the iron pad and substitute it into the second linear equation to calculate the depth of the top surface of the iron pad. ; Based on the type of iron pad, the preset thickness of the iron pad is obtained, and the difference between the top surface depth of the iron pad and the preset thickness is calculated to obtain the bottom surface depth of the iron pad. .

4. The fastener pad model identification system according to claim 3, characterized in that, Furthermore, the thickness of the fastener pad is derived to include: The total thickness of the fastener pad is calculated using the following formula: This refers to the total thickness of the fastener pad. The thickness of the shim on the fastener pad is calculated using the following formula: The thickness of the pad on the fastener pad; The thickness of the lower shim plate of the fastener shim plate is calculated using the following formula: This refers to the thickness of the underside pad of the fastener pad.

5. A fastener pad model identification system according to claim 4, characterized in that, Complete the identification of fastener pad model, including: The total thickness of the fastener pad is calculated. Thickness of the pad on the fastener pad and the thickness of the fastener pad plate under the pad plate The model of the fastener pad is identified by comparing it with the corresponding standard size in the preset database.

6. The fastener pad model identification system according to claim 1, characterized in that, The image acquisition module is specifically: The image acquisition module consists of one line structured light 3D camera, positioned directly above the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, two line structured light 3D cameras are used. Alternatively, the image acquisition module may consist of two line structured light 3D cameras, symmetrically arranged above the fastener areas on both sides of the rail, to image the fastener areas on both sides of the rail. To achieve recognition of the rail areas on both sides, four line structured light 3D cameras may be used.

7. A fastener pad model identification system according to claim 6, characterized in that, Also includes: The line structured light 3D camera includes a line structured light component and a 3D camera. The line structured light measurement algorithm built into the 3D camera is used to obtain the depth image of the fastener. The depth measurement accuracy of the line structured light 3D camera is no less than 0.1 mm, and the horizontal measurement accuracy is no less than 0.1 mm / pixel.

8. A fastener pad model identification system according to claim 1, characterized in that, Preprocessing and fastener assembly segmentation of the corrected fastener depth image include: The corrected fastener image is first subjected to linear grayscale stretching and then standardized. A segmentation model is then used to segment the fastener components in the preprocessed fastener image. Linear grayscale stretching is performed using a linear grayscale stretching formula, specifically as follows: This is a diagram showing the result after linear stretching. This is the corrected fastener depth diagram. The maximum value in the corrected fastener depth diagram. To obtain the minimum value in the corrected fastener depth diagram, This represents the maximum value in the fastener drawing after grayscale stretching. This is the minimum value in the fastener drawing after grayscale stretching; Standardization is performed using a standardized formula, specifically as follows: in, The input data before standardization. For standardized data, The mean of all image data for ballast track is given. The standard deviation of all image data for ballast track; The segmentation model used is the SINet neural network segmentation model.

9. A method for identifying the model of a fastener pad, employing the fastener pad model identification system as described in claim 1, characterized in that, include: Collect images of the fastener depth on both sides of the rail; Based on the height of the sleeper support platform, the acquired fastener depth image is height-corrected, and the corrected fastener depth image is output. The corrected fastener depth image is preprocessed and the fastener assembly is segmented to obtain the fastener assembly segmentation image; Based on the corrected fastener depth image and the segmented image of the fastener assembly, the pixel values ​​of key areas are extracted and depth parameters are fitted and calculated, thereby deriving the thickness of the fastener pad. The key areas include the effective area of ​​the rail base, the effective area of ​​the sleeper support platform, and the effective area of ​​the iron pad. The depth parameters include the top surface depth of the rail base, the bottom surface depth of the rail base, the plane depth of the support platform, the top surface depth of the iron pad, and the bottom surface depth of the iron pad. The derived fastener pad thickness is compared with the corresponding standard size in the preset database to complete the fastener pad model identification.

Images