Steel rail missed seam detection method and system based on machine vision

Abstract

The invention relates to the technical field of railway track detection, in particular to a steel rail blind seam detection method and system based on machine vision, and the method comprises the steps of track pair image acquisition, edge track extraction, seam feature quantification, detection method selection and seam width analysis. According to the method, an optimal detection mode is quickly matched, efficient and accurate gap width evaluation is realized in a complex environment, and the system comprises a track pair image acquisition module, an edge track extraction module, a gap feature quantification module, a detection method selection module and a gap width analysis module. According to the invention, the problems of low detection efficiency, insufficient precision and poor environmental adaptability in the prior art are solved, and the accuracy and timeliness of track butt joint gap width detection are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of railway track detection, and more specifically, to a method and system for detecting blind joints of steel rails based on machine vision. Background Art

[0002] In the daily maintenance and safety guarantee work of the track system, the detection of the gap width at the track joint is a crucial task. The gap width at the track joint needs to be strictly controlled within a specified range, which plays a decisive role in ensuring the smoothness, safety of train operation, and the long-term reliability of the track system. If the gap is too wide, the train will experience severe bumps and vibrations during driving, which will not only reduce the comfort of passengers but also cause additional wear and tear on vehicle components and increase the risk of safety accidents such as derailment. If the gap is too narrow, it may cause track deformation and extrusion damage due to thermal expansion and contraction.

[0003] Currently, various technical means have been developed for detecting the gap width at the track joint. Common ones include line-line detection, point-plane detection, and three-dimensional feature - thermal imaging detection, etc. The line-line detection method obtains the position information of two lines at the track joint through a specific sensor, and then calculates the gap width. This method is relatively simple to operate, but it is easily interfered by factors such as uneven track surface and stains, resulting in poor detection accuracy. Point-plane detection is based on the combination of line laser and camera to obtain the point-plane data at the track joint, and then calculates the gap width through a complex algorithm. It has high requirements for the installation accuracy of equipment, and the data processing process takes a long time, making it difficult to meet the needs of rapid detection. Three-dimensional feature - thermal imaging detection uses thermal imaging technology to obtain the temperature distribution image at the track joint, and infers the gap width by analyzing the three-dimensional features reflected by the temperature difference. This method can detect hidden defects to a certain extent, but the equipment cost is high, the detection efficiency is low, and it is relatively sensitive to environmental temperature changes.

[0004] When faced with an image of a track joint, how to quickly and accurately select a suitable method from the above detection methods to judge the gap width has become a key problem to be solved urgently. The existing technology lacks systematic guidance in this regard, and detection personnel often make selections based on experience. This not only easily leads to improper detection methods, affecting the accuracy and timeliness of detection results, but also may waste a lot of time and resources due to repeated attempts of different methods. Therefore, there is an urgent need for an intelligent detection method selection method based on image features, which can quickly and accurately match the most suitable detection technology according to the characteristics of the track joint image, so as to achieve efficient and accurate detection of the gap width at the track joint. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technologies, the purpose of the present invention is to provide a method and system for detecting blind gaps in steel rails based on machine vision. This kind of blind gap detection combines multi-dimensional image feature analysis with an intelligent selection strategy to solve the problems of low detection efficiency, insufficient accuracy, and poor environmental adaptability in detecting the gap width at the rail docking in the existing technologies. This method and system can quickly match the optimal detection method according to the specific features of the image to be analyzed, and achieve efficient and accurate gap width evaluation in complex environments.

[0006] To achieve the above purpose, the present invention provides the following technical solutions: A method for detecting blind gaps in steel rails based on machine vision, comprising the following steps: Rail pair image acquisition step: Using a vision camera installed at the bottom of the train to take an image of the rail docking as the image to be analyzed, and simultaneously recording the surface temperature of the rail docking end. Edge trajectory extraction step: In the image to be analyzed, outlining the gap edge trajectory of the rail docking end through an edge detection algorithm, and judging the clarity of the edge trajectory according to the distribution density of edge points. Gap feature quantification step: Calculating the gray difference or normalized contrast between the gap area and the background area in the image to be analyzed as the gap contrast, analyzing the cleanliness of the gap surface, and generating a multi-dimensional feature vector in combination with the illumination intensity, surface temperature, and gap contrast. Detection method selection step: Inputting the multi-dimensional feature vector into a preset analysis method selection formula, and outputting a judgment index, which is used to select the optimal detection method among the line-line detection method, point-plane detection method, and three-dimensional feature-thermal imaging map detection method. Gap width analysis step: Performing the gap width analysis task according to the optimal detection method determined in the detection method selection step.

[0007] Further, the edge trajectory extraction step includes an edge point screening strategy, and the edge point screening strategy includes an edge point density calculation step and an edge point continuity judgment step. The edge point density calculation step: In the image to be analyzed, counting the number of edge points of the gap edge trajectory in pixels, and calculating the edge point density per unit length. The edge point continuity judgment step: Judging the continuity of the edge trajectory according to the distance threshold between edge points. If the continuity is lower than the preset threshold, marking the edge trajectory as a fuzzy trajectory.

[0008] Further, the gap feature quantification step further includes a surface cleanliness evaluation strategy. The surface cleanliness evaluation strategy calculates the stain coverage rate accumulated on the surface area of the gap by analyzing the texture features and brightness distribution of the gap area in the image to be analyzed, and generates a surface cleanliness score by combining the stain coverage rate and the gap contrast ratio.

[0009] Furthermore, it also includes a real-time monitoring step. Start the real-time monitoring timer. If the gap width analysis task is not completed within the maximum allowable processing time, switch to another detection method to re-execute the analysis task.

[0010] Furthermore, the real-time monitoring step includes a method switching strategy, and the method switching strategy includes a time threshold setting step and a switching analysis method step. In the time threshold setting step, analyze the maximum processing time of each detection method based on historical data and set the maximum allowable processing time. In the switching analysis method step, select the next detection method according to the proximity between the judgment value and the value range of the detection method.

[0011] Furthermore, the gap width analysis step includes an analysis result verification strategy, and the analysis result verification strategy includes a repeated measurement step and an error correction step. In the repeated measurement step, measure the gap width at the docking part of the same track multiple times, record the results of each measurement and calculate the average value. In the error correction step, judge the stability of the measurement according to the standard deviation of the multiple measurement results. If the standard deviation exceeds the preset threshold, re-execute the measurement task or adjust the acquisition angle of the vision camera.

[0012] Furthermore, the detection method selection step includes an extreme condition trigger strategy. The extreme condition trigger strategy includes that when the gap area is blocked by stains and has the same color as the rusty area on the surface background area of the track docking end, directly trigger an ignore instruction and use a three-dimensional feature - thermal imaging detection method for analysis.

[0013] Furthermore, in the gap width analysis step, analyze two or more consecutive images to be analyzed to judge the change in the gap width, compare it with the preset change threshold, and output a normal instruction or an abnormal instruction according to the comparison result.

[0014] Furthermore, the edge trajectory extraction step includes an image preprocessing strategy, and the image preprocessing strategy includes intercepting the area at the track docking part in the image to be analyzed through an object detection algorithm and retaining track section A and track section B on both sides of the gap.

[0015] A rail blind gap detection system based on machine vision includes: Rail-to-image acquisition module, which uses a vision camera installed at the bottom of the train to capture an image of the rail docking area as the image to be analyzed, and simultaneously records the surface temperature of the rail docking end; Edge trajectory extraction module, which outlines the gap edge trajectory of the rail docking end in the image to be analyzed through an edge detection algorithm, and judges the clarity of the edge trajectory according to the distribution density of edge points; Gap feature quantification module, which calculates the gray difference or normalized contrast between the gap area and the background area in the image to be analyzed as the gap contrast, analyzes the cleanliness of the gap surface, and generates a multi-dimensional feature vector by combining the light intensity, surface temperature, and gap contrast; Detection method selection module, which inputs the multi-dimensional feature vector into a preset analysis method selection formula and outputs a judgment index, and this judgment index is used to select the optimal detection method among the online line detection method, point-plane detection method, and three-dimensional feature-thermal imaging map detection method; Gap width analysis module, which performs the gap width analysis task according to the optimal detection method determined in the detection method selection step.

[0016] Advantages of the present invention: Through multi-dimensional image feature analysis combined with an intelligent selection strategy, the present invention solves the problems of low detection efficiency, insufficient accuracy, and poor environmental adaptability in the detection of the gap width at the rail docking in the prior art, that is, it can quickly match the optimal detection method according to the specific features of the image to be analyzed, and realize efficient and accurate gap width evaluation in a complex environment. Specifically, an edge detection algorithm is combined with an edge point screening strategy, and through density calculation and continuity judgment, the gap edge trajectory is accurately outlined, and fuzzy trajectories are effectively identified, improving the accuracy of edge extraction; when quantifying the gap features, not only the gray difference and contrast are considered, but also the surface cleanliness evaluation is introduced, the stain coverage rate is analyzed through texture and brightness, and a multi-dimensional feature vector is generated by combining multiple parameters, laying a foundation for the intelligent selection of detection methods. Description of the Drawings

[0017] Figure 1 is the overall flowchart of the present invention; Figure 2 is the top view collected vertically downward by the vision camera of the present invention; Figure 3 is the axonometric view collected by the vision camera of the present invention; Figure 4 is the system fluid diagram of the present invention. Detailed Implementation Modes

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

[0019] When faced with an image of a track docking area, how to quickly and accurately select a suitable method from the above detection methods to judge the gap width has become a key problem to be solved urgently. The prior art lacks systematic guidance in this regard, and inspectors often make selections based on experience. This not only easily leads to improper detection methods, affecting the accuracy and timeliness of detection results, but also may waste a large amount of time and resources due to repeated attempts of different methods. Therefore, there is an urgent need for an intelligent detection method selection method based on image features that can quickly and accurately match the most suitable detection technology according to the characteristics of the track docking area image, so as to achieve efficient and accurate detection of the gap width at the track docking area. Therefore, the present invention designs this method and system for detecting blind gaps in steel rails based on machine vision, in combination with the attached Figures 1-3 As shown in the figure, it includes the following steps: Track pair image acquisition step: Use a vision camera installed at the bottom of the train to capture an image of the track docking area as the image to be analyzed. At the same time, use a temperature sensor or thermal imaging to synchronously record the surface temperature of the track docking end. In the initial state, as Figure 2 shown, the installation position of the vision camera needs to ensure that its lens optical axis is perpendicular to the track docking surface. The vision camera can be adjusted, and its pitch angle 4 and yaw angle 5 are adjustable, with the adjustment ranges being ±5° and ±3° respectively, and precise control is achieved through a stepper motor drive to reduce image distortion caused by shooting angle deviation. The temperature sensor or thermal imaging and the vision camera are fixed on the same bracket, and the distance between them does not exceed 10 cm to ensure the time synchronization of the surface temperature data and image acquisition; Edge trajectory extraction step: Based on a convolutional neural network model of deep learning, roughly locate the area of the track docking area, generate candidate boxes, and then screen out the optimal candidate boxes through the non-maximum suppression algorithm. Subsequently, perform refined segmentation on the image within the candidate boxes, retaining the complete structural information of track section A and track section B. During this process, calculate the illumination intensity in the image to be analyzed as an important reference parameter for subsequent analysis. The calculation of the illumination intensity is based on the overall gray value distribution of the image and is obtained by statistically calculating the average value and standard deviation of the pixel brightness values. This process provides high-quality basic data for subsequent edge trajectory extraction and feature parameter extraction; In the image to be analyzed, the edge trajectory of the gap at the orbital docking end is outlined by an edge detection algorithm (Canny operator), and the clarity of the edge trajectory is judged according to the distribution density of edge points. Specifically, the edge trajectory extraction process includes two links: edge point density calculation and continuity judgment. In the edge point density calculation link, the number of edge points on the edge trajectory of the gap is counted in pixels, and the edge point density per unit length is calculated. In the continuity judgment link, the continuity of the edge trajectory is judged according to the distance threshold between edge points. In the subjective judgment stage, if the continuity is lower than the preset threshold, the edge trajectory is marked as a fuzzy trajectory. Among them, the edge trajectory continuity is based on the edge point distribution density and the distance threshold, and the formula configuration is as follows: , where, is the edge trajectory continuity, is the number of edge points (the countable pixel points obtained from the edge detection algorithm), is the average distance between points (calculated as the total length of the trajectory divided by - 1), is the standard deviation of the distance between points (measuring the distance dispersion), is the distance between the i-th points (the Euclidean distance between adjacent edge points), is the distance threshold (a preset parameter for judging the continuity interruption), is the density scaling factor (a preset constant for controlling the contribution of density to continuity), is the penalty coefficient (a preset constant for amplifying the influence of discontinuity), is the Heaviside step function (defined as , filtering the distance exceeding the threshold), using the normalized coefficient of variation ( reflecting the density uniformity), the exponential term penalizes the points with distances exceeding the threshold ( summarizing and counting the number of discontinuous points), and the value range is , close to 1 indicating high continuity (clear edge), and close to 0 indicating low continuity (fuzzy or interrupted edge).

[0020] The steps for quantifying the gap feature are as follows: calculate the gray difference or normalized contrast between the gap area and the background area in the image to be analyzed as the gap contrast. Among them, the formula configuration of the gap contrast is as follows: , where, is the gap contrast, and the value range is , close to 1 indicating high contrast (the gap is obvious), and close to 0 indicating low contrast (the gap is fuzzy), is the average gray level of the gap area (calculated from the gray values of the image pixels), is the average gray value of the background area (gray value mean of the non-gap area), is the standard deviation of the gray value in the gap area (measuring the gray value fluctuation), is the standard deviation of the gray value of the background area, is the contrast gain parameter (preset constant), is the area of the gap area (number of pixels), is the integral domain of the gap area (image coordinate range), is the pixel point 's gray value, is the hyperbolic tangent function (normalizing the contrast to [-1, 1], taking the absolute value and then mapping to [0, 1]), is the double integral (calculating the average absolute deviation of the area).

[0021] Analyze the cleanliness of the gap surface. Specifically, it includes a surface cleanliness evaluation strategy. By analyzing the texture features and brightness distribution of the gap area in the image to be analyzed, calculate the stain coverage rate accumulated on the gap surface, and generate a surface cleanliness score by combining the stain coverage rate and the gap contrast. The surface cleanliness scoring algorithm is configured as: , where, is the surface cleanliness score of the gap, and the value range , close to 1 indicates high cleanliness, and close to 0 indicates low cleanliness, is the stain coverage rate (stain pixel area ratio, calculated from image segmentation), is the texture entropy (calculated based on the gray level co-occurrence matrix, , measuring the texture complexity, is the stain sensitivity coefficient (preset constant), is the texture scaling factor (preset constant), is the contrast coupling parameter (preset constant), is the gap contrast, is the exponential function (suppressing the influence of stains), is the Gaussian error function.

[0022] Generate a multi-dimensional feature vector from the edge trajectory clarity, gap surface cleanliness, light intensity, surface temperature, and gap contrast; Detection method selection step. Input the multi-dimensional feature vector into a preset analysis method selection formula, and output a judgment index. The analysis method selection formula is configured as: , where, is the judgment index, is the edge trajectory continuity, is the gap contrast, is the cleanliness score of the gap surface, is the light intensity, is the surface temperature, is the feature weight parameter (preset constant), is the integral offset parameter (preset constant), is the integral variable (dummy variable), is the sigmoid function (defined as , normalized to [0, 1]), is the Gaussian integral (introducing noise robustness and filtering outliers).

[0023] This judgment index is used to select the optimal detection method among the online line detection method, the point-plane detection method, and the three-dimensional feature-thermal imaging map detection method. Among them, the line-line detection obtains the position information of two lines at the orbital docking, calculates the distance between the two lines as the gap width, and is applicable to scenarios where the orbital surface is flat and there are no obvious stains, etc. It is easily affected by factors such as uneven orbital surface, stains, and uneven illumination, resulting in a decrease in detection accuracy. For example, when there are rust or oil stains on the orbital surface, the line position may not be accurately captured, causing measurement errors. Among them, the point-plane detection method projects a line onto the orbital docking to form a light strip, the camera collects the point-plane data of the light strip, and then calculates the gap width through a complex algorithm. The detection accuracy is relatively high, but the installation position accuracy requirements of sensors such as cameras are high, and the data processing process takes a long time. Among them, the three-dimensional feature-thermal imaging map detection method uses thermal imaging technology to obtain the temperature distribution image at the orbital docking, combines three-dimensional reconstruction algorithms, analyzes the three-dimensional structural features corresponding to the temperature differences, and infers the gap width. It can penetrate some stains or rust, and the detection accuracy is high, but it is sensitive to the ambient temperature and is applicable to extreme conditions (such as the gap being covered by stains and the color being confused with the background) or scenarios requiring in-depth structural analysis (triggering extreme condition strategies).

[0024] The data examples are as follows: 1. Calculation of edge trajectory continuity Number of edge points = 100 (edge points), average inter-point distance = 2.0 (pixels), standard deviation of inter-point distance = 0.5, Sequence ( = 1.8, = 2.1,...), distance threshold = 2.5, density scaling factor = 1.0, penalty coefficient = 0.1; Calculation: (assuming 10 points exceed the threshold, ), then , that is 。

[0025] 2. Gap Contrast Calculation Average gray value of the gap area = 50, average gray value of the background area = 10, standard deviation of the gray value in the gap area = 15, standard deviation of the gray value in the background area = 0.5, contrast gain parameter = 200 (pixels), gray value of pixel point = 200 (pixels), pixel point The gray value of Example: If the gray value in the gap area is uniformly 50, then ; Calculation: , integral term = 10000 / 200 = 50, then , (after normalization , assuming the maximum gray difference is 100).

[0026] 3. Surface Cleanliness Score Calculation Stain coverage rate = 0.2 (20% stains), texture entropy = 1.5, stain sensitivity coefficient = 1.0, texture scaling factor = 1.0, contrast coupling parameter = 0.5, gap contrast = 0.5, Calculation: [[ID={}53]] 。

[0027] 4. Analysis Method Selection Calculation [[ID={}62]]Edge trace continuity , gap contrast , gap surface cleanliness score , light intensity = 80 (lux), surface temperature = 30 °C, ; Calculation: Linear part = , integral is approximated as The Gaussian expectation of (numerical integration), the result (due to the large linear part, sigmoid saturation), and the initial rule: Indicates that the line-line detection method is optimal (high edge clarity, low temperature), Indicates that the line-plane detection method is optimal (medium parameters), Note: There are some curly braces in the translation where the original text seems to have some formatting or content that is not fully clear. I've tried to translate as accurately as possible based on the given text. If there are specific instructions or corrections for those parts, it would be helpful for a more precise translation.Indicates that the three-dimensional feature - thermal imaging detection method is optimal (low edge sharpness, high temperature, or low cleanliness), and the result of the embodiment is , indicating that the three-dimensional feature - thermal imaging detection method is optimal (due to low continuity and high temperature = 30°C).

[0028] Gap width analysis step, perform the gap width analysis task according to the optimal detection method determined by the detection method selection step.

[0029] It also includes a real-time monitoring step. When starting to execute a certain detection method, start the timing real-time monitoring timer. If the gap width analysis task is not completed within the maximum allowable processing time, switch to another detection method to re-execute the analysis task. The switching basis is determined by the method switching strategy. The method switching strategy includes a time threshold setting step and a switching analysis method step. To make the judgment value closer to the value range corresponding to the detection method, the specific selection rules are as follows: If the judgment value is close to the value range of the line-line detection method, give priority to switching to the line-line detection; if the judgment value is close to the value range of the point-plane detection method, give priority to switching to the point-plane detection; if the judgment value is close to the value range of the three-dimensional feature - thermal imaging detection method, give priority to switching to the three-dimensional feature - thermal imaging detection; the timer is tightly coupled with the execution logic of the detection method to ensure the timeliness and accuracy of the switching process.

[0030] The gap width analysis step includes an analysis result verification strategy. The analysis result verification strategy includes a repeated measurement step and an error correction step to ensure the accuracy and stability of the gap width detection result, where Repeated measurement step, perform more than 2 independent measurements on the gap width at the same track docking. For example, the vision cameras at the bottom of the second carriage and the third carriage collect images of the same track docking, record the results of each measurement, and calculate the average value; Error correction step, judge the stability of the measurement according to the standard deviation of the multiple measurement results. The calculation formula for the standard deviation of the measurement results is: , where is the standard deviation, is the number of measurements, is the result of the i-th measurement, is the mean of the measurement results. If the standard deviation exceeds the preset threshold, re-execute the measurement task, that is, detect the same position again, eliminate accidental interferences (such as sudden changes in light, image blurring caused by vibration), or adjust the acquisition angle of the vision camera. Change from the original vertical acquisition of one upper surface to the oblique acquisition of the upper surface and the side surface, such as a pitch angle of ±5° and a yaw angle of ±3°, to obtain the side structure information of the gap and avoid feature occlusion caused by a single shooting angle, that is, as Figure 3The image collected from the side is shown, and the line-line detection method should be selected for the gap width detection for this figure.

[0031] The detection method selection step includes an extreme condition trigger strategy. The extreme condition trigger strategy includes that when the gap is blocked by stains and the surface background area at the rail butt joint shows rust, the analysis method selection formula is ignored, and the three-dimensional feature-thermal imaging map detection method is directly called for analysis. The judgment of the gap being blocked by stains is confirmed by analyzing the color consistency between the background area and the gap area through the color histogram. If the overlap rate of the color histograms of the two exceeds 90%, the three-dimensional feature-thermal imaging map detection method is triggered. The three-dimensional feature-thermal imaging map detection method combines thermal imaging technology and three-dimensional reconstruction algorithms, and can achieve accurate evaluation of the gap width in a complex environment.

[0032] The gap width analysis step analyzes two or more consecutive images to be analyzed to judge the width change of the gap, and compares it with a preset change threshold, and outputs a normal instruction or an abnormal instruction according to the comparison result. For abnormal dynamic gaps, such as the instantaneous expansion of the gap caused by the loosening of the rail fasteners, the time window for early detection of potential hazards can reach several hours to several days compared with traditional static detection (one inspection per day).

[0033] A detection system is correspondingly designed according to the rail blind gap detection method based on machine vision, such as Figure 4 shown. The system includes an image acquisition module installed at a position near the rail butt joint at the bottom of the train, which is used to capture the image to be analyzed at the rail butt joint and synchronously record the surface temperature of the rail butt end. The edge trajectory extraction module is connected to the image acquisition module. After receiving the image to be analyzed, it extracts the edge trajectory of the gap at the rail butt end through the edge detection algorithm, and judges the clarity of the edge trajectory according to the distribution density of the edge points. The gap feature quantization module is connected to the edge trajectory extraction module, calculates the gray difference or normalized contrast between the gap area and the background area as the gap contrast, and generates a multi-dimensional feature vector by combining the light intensity, surface temperature and gap contrast. The detection method selection module receives the multi-dimensional feature vector from the gap feature quantization module, outputs a judgment index through a preset analysis method selection formula, and selects the optimal detection method according to the judgment index. The gap width analysis module is connected to the detection method selection module, executes the selected detection method to complete the gap width analysis task, and starts a real-time monitoring timer. The real-time monitoring module is connected to the gap width analysis module, monitors the execution time of the analysis task, and switches to another detection method to re-execute the analysis task in case of timeout.

[0034] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A method for detecting blind joints of steel rails based on machine vision, characterized in that: It includes the following steps: Track pair image acquisition step: Use a vision camera installed at the bottom of the train to capture an image of the track docking area as the image to be analyzed, and simultaneously record the surface temperature of the track docking end; Edge trajectory extraction step: In the image to be analyzed, outline the edge trajectory of the gap at the track docking end through an edge detection algorithm, and judge the clarity of the edge trajectory according to the distribution density of edge points; Gap feature quantification step: Calculate the gray difference or normalized contrast between the gap area and the background area in the image to be analyzed as the gap contrast, analyze the cleanliness of the gap surface, and generate a multi-dimensional feature vector in combination with the light intensity, surface temperature, and gap contrast; Detection method selection step: Input the multi-dimensional feature vector into a preset analysis method selection formula, and output a judgment index, which is used to select the optimal detection method among the online line detection method, point-plane detection method, and three-dimensional feature-thermal imaging map detection method; Gap width analysis step: Perform the gap width analysis task according to the optimal detection method determined in the detection method selection step.

2. The method for detecting blind joints of steel rails based on machine vision according to claim 1, wherein: The edge trajectory extraction step includes an edge point screening strategy, and the edge point screening strategy includes an edge point density calculation step and an edge point continuity judgment step. The edge point density calculation step: In the image to be analyzed, count the number of edge points of the gap edge trajectory in pixels, and calculate the edge point density per unit length; The edge point continuity judgment step: Judge the continuity of the edge trajectory according to the distance threshold between edge points. If the continuity is lower than the preset threshold, mark the edge trajectory as a fuzzy trajectory.

3. The method for detecting blind joints of steel rails based on machine vision according to claim 2, wherein: The gap feature quantification step also includes a surface cleanliness evaluation strategy. The surface cleanliness evaluation strategy: Analyze the texture features and brightness distribution of the gap area in the image to be analyzed, calculate the stain coverage rate accumulated on the gap surface, and generate a surface cleanliness score in combination with the stain coverage rate and the gap contrast.

4. The method for detecting blind joints of steel rails based on machine vision according to claim 1 or 3, characterized in that: It also includes a real-time monitoring step: Start a real-time monitoring timer. If the gap width analysis task is not completed within the maximum allowable processing time, switch to another detection method to re-execute the analysis task.

5. The method for detecting blind joints of steel rails based on machine vision according to claim 4, wherein: The real-time monitoring step includes a method switching strategy, and the method switching strategy includes a time threshold setting step and a switching analysis method step. The time threshold setting step: Analyze the maximum processing time of each detection method based on historical data, and set the maximum allowable processing time; The switching analysis method step: Select the next detection method according to the proximity of the judgment value to the value range of the detection method.

6. The method for detecting blind joints of steel rails based on machine vision according to claim 5, wherein: The gap width analysis step includes an analysis result verification strategy, and the analysis result verification strategy includes a repeated measurement step and an error correction step. The repeated measurement step: Measure the gap width at the same track docking area multiple times, record the results of each measurement, and calculate the average value; The error correction step: Judge the stability of the measurement according to the standard deviation of the multiple measurement results. If the standard deviation exceeds the preset threshold, re-execute the measurement task or adjust the acquisition angle of the vision camera.

7. The method for detecting blind joints of steel rails based on machine vision according to claim 1, wherein: In the detection method selection step, an extreme condition triggering strategy is included. The extreme condition triggering strategy includes that when the gap area is blocked by stains and has the same color as the rusty background area on the surface of the track docking end, an ignore instruction is directly triggered, and a three-dimensional feature-thermal imaging detection method is used for analysis.

8. The method for detecting blind joints of steel rails based on machine vision according to claim 1, characterized in that: In the gap width analysis step, two or more consecutive images to be analyzed are analyzed to judge the width change of the gap, and compared with a preset change threshold, and a normal instruction or an abnormal instruction is output according to the comparison result.

9. The method for detecting blind joints of steel rails based on machine vision according to claim 1, wherein: The edge trajectory extraction step includes an image preprocessing strategy. The image preprocessing strategy includes intercepting the track docking area in the image to be analyzed through an object detection algorithm, and retaining the track section A and track section B on both sides of the gap.

10. A rail blind joint detection system based on machine vision, characterized in that: It includes: A track pair image acquisition module, which uses a vision camera installed at the bottom of the train to capture an image of the track docking as the image to be analyzed, and synchronously records the surface temperature of the track docking end; An edge trajectory extraction module, which outlines the gap edge trajectory of the track docking end in the image to be analyzed through an edge detection algorithm, and judges the clarity of the edge trajectory according to the edge point distribution density; A gap feature quantification module, which calculates the gray difference or normalized contrast between the gap area and the background area in the image to be analyzed as the gap contrast, analyzes the cleanliness of the gap surface, and generates a multi-dimensional feature vector in combination with the light intensity, surface temperature and gap contrast; A detection method selection module, which inputs the multi-dimensional feature vector into a preset analysis method selection formula and outputs a judgment index, which is used to select the optimal detection method among the online line detection method, the point-plane detection method and the three-dimensional feature-thermal imaging detection method; A gap width analysis module, which executes the gap width analysis task according to the optimal detection method determined in the detection method selection step.

Images