A method and system for real-time monitoring of a main cable strand of a suspension bridge

Abstract

The application discloses a kind of suspension bridge main cable strand drum silk real-time monitoring method and system, it is related to bridge construction monitoring technical field.The monitoring point of the strand traction path is equipped with monitoring camera, and real-time video stream containing strand is collected;Each frame image in the video stream is subjected to image enhancement processing;Based on the image after enhancement processing, in the preset strand monitoring area, the color feature of the main body of strand is extracted, and strand color reference model is established;According to the color reference model, color distance segmentation is carried out on subsequent frame image, and the segmentation mask of the main body of strand is generated;Contour extraction and analysis are carried out on segmentation mask, and whether the strand occurs drum silk abnormality is judged based on the contour condition extracted.The application realizes the automatic, all-weather, quantifiable real-time monitoring of main cable strand drum silk phenomenon, overcomes the shortcomings that artificial monitoring is strong in subjectivity and existing visual technology has poor environmental adaptability, and improves construction safety and quality control level.

Description

Technical Field

[0001] This invention relates to the field of construction technology for main cable strands of suspension bridges, specifically to a method and system for real-time monitoring of the taut wires of main cable strands of suspension bridges. Background Technology

[0002] Suspension bridges are a type of bridge with large spans and complex structures. The erection of the main cable is one of the key steps in the construction of suspension bridges. The main cable construction mainly adopts the PPWS method, which includes six major construction procedures: cable strand traction, cable strand lateral movement, cable strand shaping, cable strand saddle insertion, cable strand alignment adjustment, and anchor span tension adjustment. The content of this patent mainly involves cable strand traction.

[0003] The main cable of a suspension bridge is composed of a large number of high-strength steel wires twisted together, typically erected in batches during construction in the form of strands. During the strand traction process, due to uneven tension, improper construction techniques, or environmental factors, some steel wires may experience bulging, meaning that a steel wire locally protrudes from or breaks off from the main strand, forming an abnormal shape separated from the main structure. This bulging phenomenon severely affects the load-bearing performance of the main cable, reducing the overall safety and service life of the bridge.

[0004] Traditional methods of monitoring cable strands require dedicated personnel to follow along and observe visually. These methods lack quantification, demand significant human resources, are heavily influenced by worker subjective judgment, have low levels of informatization and digitization, and result in highly subjective outcomes. While some existing technologies attempt to use visual methods for monitoring, they typically process the entire image globally without targeted preprocessing or dynamic Region of Interest (ROI) focusing mechanisms. This makes the algorithms susceptible to the complex and variable lighting conditions at the construction site (such as sunshine / rain, day / night), background interference (such as bridge towers and the sky), and the movement of the cable strands themselves, making it difficult to achieve stable and reliable real-time identification of cable strands. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a real-time monitoring method and system for the main cable strands of suspension bridges that can adapt to complex construction environments, achieve automation, high precision and quantification, so as to solve the problems of strong subjectivity, low efficiency and high risk of traditional manual monitoring, as well as the poor robustness and inability to quantify in real time of existing visual technology.

[0006] The technical solution of this application is: a method for real-time monitoring of the strand drum wires of the main cable of a suspension bridge, comprising: Surveillance cameras are deployed at monitoring points along the cable strand traction path to collect video streams containing the cable strand in real time; Image enhancement processing is performed on each frame of the video stream; Based on the enhanced image, the color features of the main body of the strand are extracted within the preset strand monitoring area, and a strand color benchmark model is established. Based on the color reference model, color distance segmentation is performed on subsequent frame images to generate a segmentation mask for the main body of the strand. The segmentation mask is subjected to contour extraction and analysis, and the extracted contour is used to determine whether the strand has a bulging abnormality.

[0007] According to the real-time monitoring method for the main cable strands of a suspension bridge provided in this application, the method for image enhancement processing of each frame of the video stream includes: performing bilateral filtering on the original image of the video stream to suppress noise and preserve edges; Apply an unsharpened mask to the filtered image to enhance edge details; The image with enhanced edge details is converted from the BGR color space to the LAB color space, and contrast-limited adaptive histogram equalization is applied to the luminance channel L to adaptively enhance local contrast. The image with enhanced local contrast is then converted to the HSV color space, the saturation channel S is linearly stretched and enhanced, and finally converted back to the BGR color space to output the enhanced image.

[0008] According to the real-time monitoring method for the main cable strand drum wire of a suspension bridge provided in this application, the method for generating the strand monitoring area includes: The object detection model is used to identify the idler rollers in the video frame, and it is assumed that the area above them is where the cable strands are located; Based on the boundary coordinates of the idler roller detection frame, the horizontal and vertical ranges are shifted upwards and expanded to automatically generate the monitoring area ROI covering the cable strands; A time smoothing strategy is used to stabilize the ROI location in the monitoring area; If a roller is not detected or the confidence level is too low in a certain frame, the monitoring region ROI of the previous frame will be used for further processing; if roller detection fails for multiple consecutive frames, an alarm will be triggered.

[0009] According to the real-time monitoring method for the main cable strands of a suspension bridge provided in this application, the target detection model is a roller detection model trained based on YOLOv8, and only the detection result of the single roller with the highest confidence in each frame is retained for ROI calculation of the monitoring area.

[0010] According to the real-time monitoring method for the main cable strand drum wires of a suspension bridge provided in this application, the method for extracting the color features of the main strand includes: During the system initialization phase, pixels are sampled from the monitoring region ROI of the first N frames of enhanced images and converted to the HSV color space; The sampled pixels are clustered using K-means and divided into K color clusters; Select the cluster with the most pixels as the main color cluster of the strand, and use the center of the cluster as the color reference of the strand. The color similarity threshold is adaptively determined based on the average distance from the pixel to the center within the cluster.

[0011] According to the real-time monitoring method for the main cable strands of a suspension bridge provided in this application, when the change in the global average brightness of the image exceeds the brightness threshold or the change in contrast exceeds the contrast threshold, K-means clustering is re-executed to update the color reference of the cable strands.

[0012] According to the real-time monitoring method for the main cable strands of a suspension bridge provided in this application, before color distance segmentation, the enhanced image is first subjected to adaptive statistical filtering noise reduction processing. The adaptive statistical filtering automatically removes discrete noise points by analyzing the statistical distribution of color values ​​in the neighborhood of each pixel.

[0013] According to the real-time monitoring method for the main cable strand filaments of a suspension bridge provided in this application, the segmentation mask is used to extract and analyze the contours and identify the main contours. If there is at least one non-main contour and the minimum Euclidean distance between the non-main contour and the main contour is greater than a set distance threshold, it is determined that the filaments are abnormal, and the position, area and separation distance parameters of the filaments are output.

[0014] According to the real-time monitoring method for the main cable strands of a suspension bridge provided in this application, the method for identifying the main body contour includes: Perform connected component analysis on the segmentation mask to extract all external contours; Calculate the area of ​​each contour and filter out contours whose area is less than a set area threshold; The filtered contours are sorted in descending order of area, and the contour with the largest area is identified as the main contour of the strand.

[0015] This application also relates to a real-time monitoring system for the strand drums of the main cable of a suspension bridge for implementing the above method, comprising: The data acquisition module is used to acquire video streams containing the cable strands in real time; The image processing module is used to perform image enhancement processing on each frame of the video stream; The color reference establishment module is used to extract the color features of the main body of the rope strand within a preset rope strand monitoring area based on the enhanced image and to establish a rope strand color reference model. The color segmentation module is used to perform color distance segmentation on subsequent frame images based on the color reference model to generate a segmentation mask for the main body of the cable strand. The judgment module is used to extract and analyze the contour of the segmentation mask, and to determine whether the main cable has a filament bulging abnormality based on the extracted contour.

[0016] The advantages of this application are as follows: 1. The real-time monitoring method for bulging strands of the main cable of the suspension bridge in this application constructs a complete automated monitoring process from image acquisition, enhancement, color modeling, segmentation to contour analysis and judgment, realizing 24 / 7 uninterrupted automated monitoring of bulging strands of the main cable, directly replacing the traditional high-altitude manual inspection, significantly reducing safety risks and labor costs, and making it possible to detect anomalies in real time, which is conducive to timely early warning and handling; The monitoring method of this application abstracts the bulging strand anomaly into the contour relationship problem between the main strand and the separated part in the image. Through the technical path of establishing a color reference model, color distance segmentation, and contour extraction and analysis, it breaks through the limitations of traditional reliance on manual visual observation or simple geometric dimension measurement, and provides a new and quantifiable computer vision solution for bulging strand detection, laying the foundation for the high precision and objectivity of the whole method; 2. This application specifies a four-level adaptive image enhancement process, including bilateral filtering, unsharpened mask, LAB space CLAHE, and HSV space saturation stretching. This four-level enhancement process precisely addresses common problems in construction site images such as blurring, noise, low contrast, and color distortion. Bilateral filtering reduces noise while protecting edges, unsharpened mask sharpens cable strand boundaries, CLAHE enhances local contrast to overcome uneven lighting, and saturation stretching strengthens the color difference between the cable strands and the background. This combined approach significantly improves image quality, creating a prerequisite for subsequent high-precision segmentation. Through multi-space conversion and processing, this scheme can more effectively cope with the challenges posed by different lighting conditions such as sunny days, cloudy days, rainy days, and nighttime. In particular, the brightness enhancement in LAB space and the saturation adjustment in HSV work together to ensure that cable strand features are stably highlighted in different environments, greatly enhancing the robustness of the entire monitoring system. 3. This application automatically generates and stabilizes the ROI (Region of Interest) by detecting idler rollers, specifying the use of the YOLOv8 model and taking the highest confidence result; it eliminates the need for extremely precise fixed calibration of the camera; by indirectly locating the cable strands through idler roller detection, and employing time smoothing and result inheritance strategies, it can effectively adapt to actual situations such as slight camera shaking and minor changes in cable strand position during traction, ensuring the continuous effectiveness and stability of the monitoring area; processing only the ROI area significantly reduces the computational overhead of irrelevant backgrounds, improving the system's real-time processing capabilities; at the same time, focusing on key areas also reduces the impact of background interference on segmentation and judgment algorithms; when idler rollers cannot be detected for multiple consecutive frames, an alarm is triggered, which can promptly detect camera obstruction, serious malfunctions, or abnormal operating conditions, improving the system's reliability and practicality; 4. In the initialization phase, this application adaptively establishes the color benchmark and similarity threshold of the strands through K-means clustering, and triggers updates when there are drastic changes in lighting. By learning the main color of the strands from the first N frames of actual images through clustering, rather than relying on fixed color values, the system can automatically adapt to the natural differences in surface color of strands of different projects, batches, and conditions, making it more versatile. The benchmark update is triggered based on changes in global brightness or contrast, allowing the color model to follow the long-term slow changes or short-term drastic changes in ambient lighting caused by sudden weather changes from morning to night, ensuring the continuous effectiveness of the color segmentation algorithm throughout the monitoring period, which is a key part of the system's high robustness. The threshold is adaptively determined based on the average distance from the pixel to the center within the cluster, which is more scientific than empirical setting. It can effectively distinguish abnormal parts such as bulging wires while ensuring that the strands are completely segmented, thus improving the accuracy of segmentation. 5. Before color segmentation, this application uses an adaptive filter based on neighborhood statistical distribution to remove discrete noise points, which is specifically used to handle discrete noise points (such as salt and pepper noise) that may be generated by vibration of the monitoring platform, electromagnetic interference or the sensor itself. Removing them before segmentation can effectively prevent noise points from being missegmented into tiny abnormal contours, greatly reducing the false alarm rate of subsequent contour analysis and improving the accuracy and reliability of the detection results. 6. This application quantifies the drum wire as a contour that is separated from the main strand and at a distance exceeding a threshold. This definition captures the most core morphological feature of the drum wire in the image—spatial separation. Compared with methods that monitor width changes or shape regularity, this is more direct and more resistant to interference. The algorithm of this application can output specific parameters such as the position, area, and separation distance of the drum wire. This completely changes the traditional subjective qualitative judgment mode that relies on experience, providing objective and quantifiable data for construction quality assessment, anomaly classification and early warning, and maintenance decisions, and realizing the standardization and scientification of the detection process. Through multi-layer logic of connected component analysis, area filtering, main body recognition, and distance determination, it can effectively distinguish between real drum wires and interference contours formed by image noise, water droplet reflection, and attachments, improving the detection rate while ensuring the specificity of the judgment (low false alarm rate). 7. This application breaks down the entire complex process into clear functional modules such as data acquisition, image processing, color benchmark establishment, color segmentation, and judgment, which reflects the rationality of the system design and the feasibility of engineering implementation, and also provides direct support for the system architecture description in the specification. Attached Figure Description

[0017] Figure 1 : Flowchart of the real-time monitoring method for the main cable strand drum of the suspension bridge in this application; Figure 2 : Schematic diagram of the main outline and secondary outline of this application. Detailed Implementation

[0018] The embodiments of this application are described in detail below, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0019] In the description of this application, it should be understood that the terms "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0020] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0021] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0022] This application relates to a real-time monitoring method for the bulging of strands in the main cable of a suspension bridge. This application constructs an intelligent monitoring scheme for the bulging of strands in the main cable of a suspension bridge, characterized by a complete technical closed loop, strong environmental adaptability, high detection accuracy, and objective quantitative results. This scheme overcomes environmental interference through multi-level adaptive image enhancement; ensures segmentation robustness through dynamic color modeling and statistical filtering; and finally achieves accurate identification and quantification of bulging anomalies through innovative contour separation analysis. This scheme transforms traditional high-risk, subjective high-altitude manual inspection into a safe, automated, objective, and data-driven modern construction quality control method, possessing significant technological advancement and practical value.

[0023] Specifically, such as Figure 1 As shown, the real-time monitoring method for the main cable strands of a suspension bridge according to this application includes the following steps performed in sequence: S1. Deploy high-definition industrial cameras at key monitoring points along the cable strand traction path (in this application, fixed monitoring cameras are installed near the cable tower saddle) to collect video streams containing the moving cable strands in real time. S2. Perform image enhancement processing on each frame of the original image in the video stream to improve image quality; S3. Based on the enhanced image, extract representative color features of the main body of the rope strand within the preset rope strand monitoring area, and establish a rope strand color benchmark model through modeling. S4. Based on the established color reference model, perform image segmentation based on color distance on each subsequent enhanced image to generate a binarized segmentation mask that only contains the main body of the strand and possible filament regions. S5. Perform contour extraction and geometric analysis on the obtained segmentation mask. Based on the number, position and spatial relationship of the extracted contours, automatically determine whether the cable strands in the current frame have a bulging abnormality, and output the judgment result and quantization parameters.

[0024] The monitoring method of this application transforms the filament bulging defect in the physical world into a foreground target separation problem in computer vision. By establishing a stable color feature model of the main strand, the strand can be accurately separated from a complex background. Under normal circumstances, the segmentation mask should present a complete connected region (the main strand); when filament bulging occurs, some wires will detach from the main cable bundle, appearing on the image as one or more independent connected regions spatially separated from the main region; by detecting and analyzing these separated contours, automatic identification of the filament bulging can be achieved.

[0025] In some embodiments of this application, the image enhancement process method described above has been optimized. Specifically, this embodiment details the specific method for enhancing each frame of the video stream; the process includes four sequentially executed sub-steps: S2.1, Bilateral Filtering A bilateral filter is applied to the original BGR image to smooth image noise while effectively preserving the edge information of the strands; in: For (x, y) Centered neighborhood window ( diameter d = 9）； For spatial Gaussian kernels, the standard deviation is... s s = 75; For color Gaussian kernel, standard deviation s r = 75; W p These are the normalization coefficients; Bilateral filtering can smooth noise while maintaining edge sharpness and avoid blurring of wire boundaries; S2.2, Unsharpened Mask An unsharpened mask algorithm is applied to the filtered image to enhance the sharpness of the strand contours and surface textures. Gaussian blur (σ=1.0) is applied to the denoised image, and then sharpening is achieved through weighted overlay: in: The image after sharpening; This is the image after bilateral filtering; The image after Gaussian blurring; This operation enhances image edges and details, making abnormal features such as filaments more prominent, and the result is truncated to prevent pixel overflow. S2.3, LAB color space local contrast enhancement The image is converted from the BGR color space to the LAB color space, and a contrast-limited adaptive histogram equalization algorithm is applied to the luminance channel (L) to adaptively improve the contrast of each local area of ​​the image. CLAHE divides the image into 16×16 blocks, performs histogram equalization on each block independently, and limits the contrast amplification factor (clip Limit = 2.0); the boundaries between adjacent blocks are smoothly transitioned using bilinear interpolation to avoid block artifacts; this method can adaptively enhance local contrast and improve the visibility of details in both dark and bright areas. S2.4, HSV color space saturation enhancement The image is converted to the HSV color space, and the saturation channel (S) is linearly stretched with an increment of 15. Saturation enhancement can amplify the color differences between different steel wire surfaces, improve the discriminative power of subsequent clustering and segmentation, and enhance color vibrancy. Finally, the image is converted back to the BGR color space for output.

[0026] The four-level enhancement process in this embodiment is specifically designed to address common problems in engineering site images; bilateral filtering utilizes the dual weights of spatial proximity and pixel value similarity to protect edges during noise reduction; the unsharpened mask enhances high-frequency edges by subtracting the blurred version of the original image; CLAHE overcomes global illumination unevenness by calculating and equalizing the histogram of local regions; and HSV spatial saturation stretching directly enhances the difference in color purity between the strands (usually metallic) and the background (sky, bridge towers, etc.).

[0027] The image processing method described in this embodiment can address issues such as noise, blurred edges, low overall contrast, and dull colors. The enhanced image exhibits more stable and prominent features under different times and weather conditions, providing high-quality, clear input for subsequent color modeling and segmentation. This is one of the fundamental guarantees of the high robustness of the entire method. CLAHE's adaptive properties enable it to cope with uneven lighting, avoiding local overexposure or underexposure that may occur during global processing.

[0028] In other embodiments of this application, the adaptive monitoring region generation method described above is optimized. Specifically, this embodiment illustrates how to dynamically determine and stabilize the image region to be analyzed; the method includes: S3.1 Idler Roller Inspection Using a pre-trained object detection model based on the YOLOv8 framework, roller (roller used to support the cable strand) is identified in each frame of the image; S3.2 ROI Calculation In each frame, only the roller detection frame with the highest confidence level is retained. Where: B is the set of all detected idler roller bounding boxes; ( () represents the coordinates of the center of the detection frame; ( () represents the width and height of the detection frame; c i Confidence score; N det The number of idlers detected; Based on the coordinates of its upper left and lower right corners, the system translates upwards (towards the negative Y-axis of the image coordinate system) by a certain number of pixels and expands horizontally by a certain range, automatically calculating a rectangular area as the tangent detection area (ROI). The cable strand is located directly above the idler roller. The monitoring area is determined by shifting the entire idler roller detection frame upwards. Let the coordinates of the four boundaries of the idler roller detection frame be: , in: The coordinates of the left boundary of the idler roller detection frame; x is the x-coordinate of the center of the idler roller detection frame; w represents the width of the idler roller detection frame; The coordinates of the right boundary of the idler roller detection frame; The coordinates of the upper boundary of the idler roller detection frame; The coordinates of the lower boundary of the idler roller detection frame; y is the ordinate of the center of the idler roller detection frame; h is the height of the idler roller detection frame; S3.3, ROI stable A moving average filter is applied to the ROI coordinates calculated from consecutive frames to smooth out positional fluctuations. To address slight camera jitter or minor adjustments in roller position, the system employs a time smoothing strategy. During the initial stabilization phase, after the system starts up, the position of the idler rollers is continuously detected for the first 10 frames. The ROI coordinates are then filtered by mean to eliminate instantaneous detection errors and obtain a stable initial monitoring area. During the tracking phase, the idler rollers are re-detected and the ROI position is updated every 30 frames (approximately 2 seconds), and smoothed using an exponentially weighted moving average. This strategy allows the ROI position to slowly follow the movement of the idler roller, which can both adapt to changes in the actual position and suppress drastic jumps caused by detection noise. S3.4, Troubleshooting If a roller is not detected in a frame or the highest confidence level is below the threshold, the valid ROI of the previous frame is used. If the detection fails for M consecutive frames (e.g., 10 frames), the system triggers an alarm and needs to check the camera for obstruction or offset.

[0029] This embodiment is based on the prior knowledge that the relative positional relationship between the cable strand and the idler roller is stable; by detecting the idler roller with a relatively fixed position and obvious characteristics, the cable strand in motion can be located indirectly and dynamically; the time smoothing strategy (moving average) can suppress ROI jitter caused by model detection fluctuations or slight lens shake; the fault handling logic ensures the continuity of the system under brief disturbances and provides timely alarms when serious anomalies occur.

[0030] This embodiment achieves intelligent focusing and dynamic fault tolerance in monitoring, reduces computational costs and interference, processes only images within the ROI, significantly reduces data processing volume, improves system real-time performance, and eliminates most interference from irrelevant backgrounds; no millimeter-level precision calibration of the camera is required, and the system can automatically track the cable strands in traction, adapting to actual construction scenarios; through confidence filtering, time smoothing, and result inheritance mechanisms, it effectively addresses single-frame false detections, missed detections, or momentary occlusions, ensuring continuous and stable monitoring; the introduced alarm mechanism further demonstrates the system's self-monitoring capabilities.

[0031] In some preferred embodiments of this application, the above-described adaptive color feature modeling and updating method is optimized. Specifically, this embodiment details the method for establishing and maintaining the Sogo color reference model, which includes two stages: initialization and updating. Initialization stage: S3.5. When the system starts running, a large number of pixels are randomly sampled from the ROI of the enhanced images in the first N frames (e.g., N=30). S3.6. Convert all sampled pixels to the HSV color space. The HSV color space is more robust to changes in lighting than the RGB color space. S3.7. Use the K-means clustering algorithm (K value can be set to 3-5, in this embodiment it is set to 3) to divide these pixels into K color clusters. K-means optimizes the clusters by iteratively making the pixels within the clusters as similar as possible and the pixels between the clusters as different as possible. S3.8. Select the cluster with the most pixels as the main color cluster for the index, and calculate the center value of all pixels in this cluster in the HSV space, such as ( H c ，S c ，V c ), as the initial reference for the strand color; S3.9 Calculate the average Euclidean distance of all pixels in the cluster to the cluster center, multiply it by a coefficient (such as 1.5), and use it as the color similarity threshold for subsequent segmentation. By appropriately relaxing the threshold, the segmentation recall rate can be improved. Update Phase: During system operation, the system monitors the global average brightness and contrast of the image in real time (e.g., calculating the mean and standard deviation of the grayscale image of the ROI region); when the change in the average brightness exceeds the brightness threshold... ΔL Or the contrast change exceeds the contrast threshold ΔC If a significant change in the lighting environment is detected, steps S3.5 to S3.9 are automatically re-executed to update the color reference and threshold.

[0032] The color distribution of the same strand under specific lighting conditions is relatively concentrated, and its dominant color can be found through unsupervised clustering; K-means can automatically distinguish the main body, shadow, highlight and background color of the strand; selecting the largest cluster ensures that the model represents the main visible part of the strand; the threshold is determined based on the statistical distance within the cluster, which is more consistent with the actual color distribution than a fixed threshold; the update mechanism enables the model to follow long-term (such as from morning to night) or drastic (such as cloud cover) changes in ambient light.

[0033] In practical applications, the update judgment can be performed once every 100 frames; the change in the average brightness can be obtained by calculating the difference between the average value of the V channel (in HSV) of the ROI region of the current frame and the reference frame when the model was built; the change in contrast can be obtained by calculating the change in the standard deviation of the grayscale image; when an update is triggered, remodeling should be performed in a background thread to avoid blocking the real-time processing flow, and the old model should be atomically replaced after the new model is built.

[0034] This embodiment ensures the long-term effectiveness of the segmentation algorithm. The system can autonomously learn the color characteristics of the currently monitored strands without the need for manual color range calibration, and can adapt to strands of different projects and materials. Based on the dynamic threshold of data distribution, the segmentation can both accommodate the natural color fluctuations on the surface of the strands and effectively distinguish abnormal areas, thus improving the accuracy of the segmentation. The introduced update mechanism is no longer a one-time calibration, but has the ability to operate stably for a long time and can cope with all-weather and multi-season light changes, which is another core pillar of the system's high robustness.

[0035] In some embodiments of this application, the adaptive statistical filtering noise reduction method described above is optimized. Specifically, this embodiment describes a preprocessing step performed on the enhanced image before color distance segmentation; specifically: for the enhanced ROI region image (BGR format), a neighborhood window of odd size (e.g., 5x5) is defined with the pixel to be processed as the center; the mean value of all pixels within this window on the three channels of BGR is calculated. m B ,m G ,m R ) and standard deviation ( s B ,s G ,s R For the center pixel, the values ​​of its B, G, and R channels are denoted as follows: p B ,p G ,p R The judgment rule is: if for any channel c, the following condition is met... |p c -m c |>k*s c (where k is a preset coefficient, such as 2.0), then the pixel is determined to be a discrete noise point; the pixel values ​​of all pixels determined to be noise points are replaced with the mean or median of the remaining pixels (non-noise points) in their neighborhood window.

[0036] Within a local neighborhood, pixels belonging to the same object (such as strands) should exhibit a certain concentrated distribution of color values; discrete noise points (such as salt and pepper noise) will deviate significantly from this distribution; by calculating local statistics and setting reasonable deviation thresholds, these abnormal pixels can be automatically identified and corrected, thereby purifying the image while maintaining the main structure of the image.

[0037] In practice, each internal pixel in the ROI image can be traversed (edge ​​pixels can be specially processed or ignored); to improve efficiency, integral image technology can be used to quickly calculate the pixel sum and sum of squares of any rectangular region, thereby accelerating the calculation of mean and standard deviation; when implementing this, attention should be paid to filling the image boundaries.

[0038] The method in this embodiment can accurately suppress noise and effectively filter out random discrete noise points introduced by sensor thermal noise, circuit interference, or transmission process. These noises are easily misjudged as tiny abnormal regions in subsequent segmentation. The filtered image is then subjected to color segmentation, resulting in a cleaner mask with fewer connected components, which significantly reduces false alarms caused by noise in subsequent contour analysis and improves the accuracy and reliability of the entire detection process.

[0039] In other embodiments of this application, the above-described contour separation anomaly detection algorithm is optimized. Specifically, this embodiment details the specific rules for judging filament anomalies based on segmentation masks; the steps include: S5.1 Contour Extraction and Preprocessing A morphological closing operation (dilation followed by erosion) is applied to the binary segmentation mask to fill small holes and smooth the boundaries, and then a contour finding algorithm is used to extract all contours. S5.2 Contour Filtering and Subject Recognition Calculate the area of ​​each contour and filter out those with an area smaller than a set threshold (e.g., 10 pixels). 2 The smallest outline of the strand (which may be residual noise) is identified; the remaining outlines are sorted by area from largest to smallest, and the outline with the largest area is identified as the main outline of the strand. S5.3 Drum wire determination and quantification Check whether there are other contours besides the main contour (called secondary contours), such as Figure 2 As shown; for each secondary contour, calculate the minimum Euclidean distance from all its pixels to the nearest pixel of the main contour; if there is at least one secondary contour whose minimum Euclidean distance is greater than a preset distance threshold (e.g., 5 pixels), then it is determined that a filament abnormality has occurred in the current frame; at the same time, record and output the position (center of the circumscribed rectangle or centroid of the contour), area, and minimum separation distance between the filament contour and the main contour.

[0040] In this embodiment, the filament is defined as an independent part that is significantly separated from the main body in space; the integrity of the outline is ensured through morphological optimization, and the area sorting is based on the physical fact that the volume of the filament part is smaller than that of the main body; the spatial distance determination directly quantifies the degree of separation, which is more accurate than simply detecting multiple outlines or shape analysis, and can effectively distinguish between closely attached objects (close distance) and the real filament (far distance).

[0041] In practice, you can use a function to calculate the distance from each pixel in the image to the main outline, and then query the value of the secondary outline pixels on the distance map and take the minimum value; or you can traverse each boundary point of the secondary outline, calculate its distance to all boundary points of the main outline, and take the global minimum value; the former is more efficient.

[0042] This embodiment can output the position, area, and distance parameters of the filament, providing objective quantitative indicators for assessing the severity of the filament damage, making quality judgments based on evidence, and achieving standardized testing; the dual rules of area filtration and distance determination can effectively eliminate the adhesion or interference from adjacent small areas caused by water droplets, stains, and reflections; the quantitative parameters help to classify and warn about the filament damage (such as minor or severe) and provide precise location guidance for subsequent repair work.

[0043] In practical applications, the real-time monitoring method for the main cable strands of suspension bridges according to this application is carried out according to the following procedure: 1. System Deployment and Initialization High-definition network cameras (with a recommended resolution of no less than 1920×1080, a frame rate of no less than 15fps, and automatic exposure and white balance functions to adapt to changes in lighting) are installed at key points (tower saddles) along the main cable strand traction path of the suspension bridge to ensure that their field of view clearly covers the cable strands and the idler rollers below. The cameras are connected to an edge computing industrial control computer deployed nearby via a network. The monitoring system software is installed in the industrial control computer. 2. Initiation and Adaptive Learning Phase After the system starts, perform the following initialization operations: a. Video stream access: The data acquisition module turns on the camera and begins to receive the real-time video stream; b. Region self-calibration: For the initial consecutive frames, after the image processing module performs basic enhancement, the idler roller detection model (YOLOv8) identifies the idler roller and calculates a stable monitoring region ROI according to the rules in the above embodiment; if stable detection cannot be achieved for multiple consecutive frames, the system alarms and prompts that the camera view should be checked. c. Color model self-establishment: After obtaining a stable ROI, the system automatically enters the color learning stage; according to the method in the above embodiment, the pixels in the first N frames of ROI are collected for K-means clustering to establish an initial ROI color benchmark model and adaptive segmentation threshold. 3. Real-time monitoring of the main loop After the system enters a stable operating state, each frame of image is processed according to the following pipeline: Step 1: Image Acquisition and Enhancement The data acquisition module captures a frame of raw image; the image processing module immediately applies the four-level adaptive image enhancement process in the above embodiment to the raw image and outputs a high-quality enhanced image. Step 2: ROI Extraction and Stabilization For the high-quality enhanced image, use the idler detection model to obtain the idler position of the current frame, and calculate the stable ROI area finally used in the current frame according to the time smoothing strategy in the above embodiments I roi ; If the detection fails, use the ROI of the previous frame Step 3: Image Purification For I roi Apply the adaptive statistical filtering in the above embodiments to eliminate possible discrete noise points and obtain a purified image Step 4: Color Segmentation The color segmentation module converts the purified image to the HSV space and reads the current valid color reference model (central value C ref and threshold T ); Calculate the Euclidean distance D between the HSV vector of each pixel in the purified image and the central value C ref ; Generate a binary segmentation mask M, where the pixels with D < T are set to 255 (foreground), otherwise set to 0 (background) Step 5: Drum Wire Analysis and Judgment The judgment module executes the contour separation degree anomaly detection algorithm in the above embodiments on the mask M Perform morphological closing operation optimization on M Extract all external contours and filter out contours with extremely small areas Identify the contour with the largest area as the main contour Traverse other contours and calculate their minimum spatial distance to the main contour Judgment logic: If there exists any other contour that satisfies the minimum spatial distance > set distance to it (for example, 10 pixels), it is determined that the drum wire is abnormal Quantitative output: If abnormal, record and output the center (position) of the circumscribed rectangle of the contour, pixel area (converted to actual area), and separation distance Step 6: Result Output and Response; Display the processing results of the current frame (original image, enhanced image, segmentation mask, contour overlay image) on the human-machine interface in real time; At the same time, draw the ROI box, main contour, and abnormal contour on the image; If it is determined as abnormal in Step 5, then Pop up a red alarm box at a prominent position on the interface to display the drum wire alarm List the detailed quantitative parameters: [Alarm time] Position: (X, Y), Drum wire area: Amm 2 , Separation distance: dmm Trigger the on-site sound and light alarm The images and data snapshots of the alarm frames are uploaded to the cloud monitoring center database via the network, and messages are pushed to the mobile apps of relevant personnel; 4. Backend adaptive maintenance While monitoring the loop in real time, the system performs maintenance tasks in parallel in the background: Color model monitoring and updating: Every certain number of frames (e.g., 300 frames, about 20 seconds), the color benchmark establishment module will check the changes in global brightness and contrast of the image; if the changes exceed the thresholds (brightness threshold ΔL, contrast threshold ΔC), a background thread will be automatically started to re-perform K-means clustering using the latest image data, and silently update the color benchmark model and thresholds to ensure that the model always adapts to the current environment. System health diagnosis: Continuously monitor the number of consecutive failed frames of idler detection, abnormal image brightness, communication status, etc.; when multiple consecutive positioning failures or other hardware failures as defined in the above embodiments occur, trigger a system-level alarm.

[0044] In addition, this application also relates to a real-time monitoring system for the main cable strands of a suspension bridge. Specifically, the system includes, in terms of hardware, a high-definition industrial camera deployed on-site, an edge computing industrial control computer or network switch, and a cloud / local server; in terms of software / logic, it includes the following functional modules: a data acquisition module, responsible for controlling camera parameters, acquiring and caching video streams; an image processing module, integrating the aforementioned image enhancement algorithm and the filtering algorithm of Example 5; a color reference establishment module, executing the aforementioned color modeling and update logic; a color segmentation module, performing color distance calculation and binarization on the image based on the color reference model and threshold; a judgment module, integrating the aforementioned contour analysis and anomaly judgment algorithms; furthermore, the system also includes a control and communication module, a human-computer interaction module (for displaying video, segmentation results, and alarm information), and a data storage module.

[0045] This system instantiates each algorithm step in the above embodiments into a functional module in the software system and deploys it on the corresponding hardware carrier. Each module interacts with data through clearly defined interfaces (input and output) and works together to complete the complete conversion process from the original video stream to the drum alarm information. The system can adopt an edge-cloud collaborative architecture, with the edge responsible for real-time processing and alarm, and the cloud responsible for data aggregation, in-depth analysis and model iterative training.

[0046] The on-site camera transmits video streams to the edge computing industrial control computer via network cable or fiber optic cable. After the system software running on the industrial control computer starts, it first initializes each module and loads the idler roller detection model. Then it enters the main loop: the data acquisition module acquires each frame of image; the image processing module enhances and reduces noise; the color segmentation module uses the model provided by the color reference module to perform segmentation; the judgment module analyzes the segmentation results and draws a conclusion; the results are displayed on the local screen through the human-machine interaction module and uploaded to the monitoring center through the control and communication module; the monitoring center can configure parameters and query the status of the edge system through the network.

[0047] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. A method for real-time monitoring of the strand diameter of the main cable of a suspension bridge, characterized in that, include: Surveillance cameras are deployed at monitoring points along the cable strand traction path to collect video streams containing the cable strand in real time; Image enhancement processing is performed on each frame of the video stream; Based on the enhanced image, the color features of the main body of the strand are extracted within the preset strand monitoring area, and a strand color benchmark model is established. Based on the color reference model, color distance segmentation is performed on subsequent frame images to generate a segmentation mask for the main body of the strand. The segmentation mask is subjected to contour extraction and analysis, and the extracted contour is used to determine whether the strand has a bulging abnormality.

2. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 1, characterized in that, The method for performing image enhancement processing on each frame of the video stream includes: performing bilateral filtering on the original image of the video stream to suppress noise and preserve edges; Apply an unsharpened mask to the filtered image to enhance edge details; The image with enhanced edge details is converted from the BGR color space to the LAB color space, and contrast-limited adaptive histogram equalization is applied to the luminance channel L to adaptively enhance local contrast. The image with enhanced local contrast is then converted to the HSV color space, the saturation channel S is linearly stretched and enhanced, and finally converted back to the BGR color space to output the enhanced image.

3. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 1, characterized in that, The methods for generating the stock detection monitoring area include: The object detection model is used to identify the idler rollers in the video frame, and it is assumed that the area above them is where the cable strands are located; Based on the boundary coordinates of the idler roller detection frame, the horizontal and vertical ranges are shifted upwards and expanded to automatically generate the monitoring area ROI covering the cable strands; A time smoothing strategy is used to stabilize the ROI location in the monitoring area; If a roller is not detected or the confidence level is too low in a certain frame, the monitoring region ROI of the previous frame will be used for further processing; if roller detection fails for multiple consecutive frames, an alarm will be triggered.

4. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 3, characterized in that, The target detection model is a roller detection model trained based on YOLOv8, and only the detection result of the single roller with the highest confidence in each frame is retained for the calculation of the ROI of the monitoring area.

5. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 1, characterized in that, The method for extracting the color features of the main body of the strand includes: During the system initialization phase, pixels are sampled from the monitoring region ROI of the first N frames of enhanced images and converted to the HSV color space; The sampled pixels are clustered using K-means and divided into K color clusters; Select the cluster with the most pixels as the main color cluster of the strand, and use the center of the cluster as the color reference of the strand. The color similarity threshold is adaptively determined based on the average distance from the pixel to the center within the cluster.

6. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 5, characterized in that, When the change in the global average brightness of the image exceeds the brightness threshold or the change in contrast exceeds the contrast threshold, K-means clustering is re-executed to update the color reference.

7. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 1, characterized in that, Before color distance segmentation, the enhanced image is first subjected to adaptive statistical filtering noise reduction processing. The adaptive statistical filtering automatically removes discrete noise points by analyzing the statistical distribution of color values ​​in the neighborhood of each pixel.

8. The method for real-time monitoring of the strands of the main cable of a suspension bridge according to claim 1, characterized in that, The segmentation mask is subjected to contour extraction and analysis to identify the main contour. If there is at least one non-main contour and the minimum Euclidean distance between the non-main contour and the main contour is greater than the set distance threshold, it is determined to be a drum wire abnormality, and the position, area and separation distance parameters of the drum wire are output.

9. A method for real-time monitoring of the strands of a suspension bridge main cable according to claim 8, characterized in that, The method for identifying the subject contour includes: Perform connected component analysis on the segmentation mask to extract all external contours; Calculate the area of ​​each contour and filter out contours whose area is less than a set area threshold; The filtered contours are sorted in descending order of area, and the contour with the largest area is identified as the main contour of the strand.

10. A real-time monitoring system for the strands of the main cable of a suspension bridge for implementing the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire video streams containing the cable strands in real time; The image processing module is used to perform image enhancement processing on each frame of the video stream; The color reference establishment module is used to extract the color features of the main body of the rope strand within a preset rope strand monitoring area based on the enhanced image and to establish a rope strand color reference model. The color segmentation module is used to perform color distance segmentation on subsequent frame images based on the color reference model to generate a segmentation mask for the main body of the cable strand. The judgment module is used to extract and analyze the contour of the segmentation mask, and determine whether the main cable has a filament bulging abnormality based on the extracted contour.

Images