Event-triggered multi-path visual inspection method and system for bridge expansion device

Abstract

The application discloses a kind of bridge expansion device multi-path visual inspection method and system based on event triggering, which comprises: multiple cameras are arranged on both sides of expansion joint, and real-time continuous images are received by parallel image acquisition threads;Event trigger signal is generated according to load change, vibration amplitude or traffic impact, key working condition images are collected, four corner points are identified using key point detection model, and perspective transformation matrix is constructed to expand expansion joint area into orthographic view;Extract multiple profiles at the preset height of the expanded image, perform grayscale, threshold segmentation and connected component analysis, estimate the pixel beam width according to the nominal width of the steel beam and calculate the scaling factor, and take the average of multiple profiles as the adaptive calibration value;Identify the steel beam from the profile and calculate the gap width between adjacent beams, only keep the gap observed by at least two profiles, convert the pixel width to millimeters according to the scaling factor, form a gap sequence and calculate the safety index, and output a safety alarm after comparing with the threshold value.

Description

Technical Field

[0001] This invention belongs to the field of bridge gap detection technology, and more specifically, relates to a multi-channel visual inspection method and system for bridge expansion joints based on event triggering. Background Technology

[0002] Bridge expansion joints are crucial components of bridges, primarily functioning to accommodate deformations in the bridge deck caused by temperature changes, vehicle loads, and impacts, ensuring the overall safety and service life of the bridge structure. Modular expansion joints (hereinafter referred to as "modular joints") are a commonly used type of bridge expansion joint. Their structure consists of a grid-like structure composed of multiple central beams and two side beams, which transfer the load to the bridge ends via supporting crossbeams. In practical applications, modular joints typically incorporate a displacement control system to ensure that each grid moves synchronously during expansion and contraction, maintaining a uniform gap distribution to accommodate the bridge's deformation requirements.

[0003] However, during long-term use, grit left by vehicles and debris washed away by rainwater can easily accumulate in the gaps, causing some grille movement to be obstructed and resulting in uneven gap distribution. In addition, factors such as vehicle impact, wear of sliding pairs, and material aging may damage or malfunction the displacement control system, which can also cause uneven gaps.

[0004] When the distribution of gaps in modular expansion joints is severely uneven, it may lead to damage or detachment of the waterstop, allowing rainwater, oil stains, and other contaminants to enter the expansion joint and reduce its durability. In severe cases, it may even cause deformation of the steel structure of the expansion joint, affecting its load-bearing capacity and threatening traffic safety. Therefore, changes in the uniformity of gaps can effectively reflect the accumulation of debris inside the expansion joint, the status of the displacement control system, and the deformation of the central beam. Effectively monitoring the uniformity of modular expansion joint gaps and analyzing its changing trends is of great significance for the operation, maintenance, and safety management of bridge expansion joints. Summary of the Invention

[0005] To address the above technical problems, this invention proposes a multi-channel visual inspection method for bridge expansion joints based on event triggering, comprising: Step 101: Install multiple cameras on both sides of the telescopic device, establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. Step 102: Generate event trigger signals based on the load changes, vibration amplitude, or traffic impact conditions of the bridge, and collect key load condition images. Automatically identify the four corner points through the key point detection model, sort the four corner points and construct a perspective transformation matrix. Expand the expansion joint area into an orthographic view through the perspective transformation matrix to obtain a geometrically consistent expanded image of the expansion joint area. Preferably, regarding the event triggering signal, this embodiment provides the following example: , Preferably, this embodiment provides the following example regarding images of key load conditions: , It should be noted that the above examples are for illustrative purposes only and are not intended to be limiting.

[0006] Step 103: Extract multiple cross-sections at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each cross-section, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, average the scaling factors obtained from all cross-sections, and obtain the adaptive scaling calibration value of the unfolded image. Step 104: Perform threshold segmentation and connected component analysis on each profile to identify steel beams. Define the blank pixel length between adjacent steel beams as the gap width. Only retain the same gap that is observed in at least two profiles as a valid gap. Replace the gap width of the valid gap with millimeter distance according to the scaling factor to generate the gap sequence corresponding to the unfolded image. Calculate multiple safety indicators for the gap sequence and compare the safety indicators with the corresponding indicator thresholds. Based on the comparison results, issue a safety alarm.

[0007] Furthermore, step 101 also includes preprocessing the real-time continuous frame image, the preprocessing operation including: eliminating image noise and increasing the contrast of the real-time continuous frame image by using Gaussian filtering and histogram equalization. The edge features of lane gaps in real-time continuous frame images are extracted using the Canny edge detection algorithm. Optimize edge continuity of real-time consecutive frame images through morphological operations; Hough transform is used to detect straight line edges, thereby improving the accuracy of edge extraction.

[0008] Furthermore, step 102 also includes: generating a timed capture signal according to a preset period, acquiring periodic high-definition images, and timestamping and queuing key load condition images and periodic high-definition images from the same camera to form a unified image input sequence.

[0009] Furthermore, step 104 includes: eliminating false beams and determining valid beams based on the expected beam width range, and defining the blank pixel length between adjacent valid beams as the seam width.

[0010] Furthermore, several safety metrics include: the maximum, minimum, average, range, and variance of the gap sequence.

[0011] Furthermore, comparing the safety indicators with the corresponding threshold values ​​includes setting a minimum and a maximum threshold value for each camera. If any safety indicator of each camera is less than the corresponding minimum threshold value or greater than the maximum threshold value, an alarm is triggered, and the alarm log is output synchronously and highlighted in the visualization interface.

[0012] Furthermore, it also includes storing all security indicators and gap sequences to generate multi-curve trend charts that evolve over time.

[0013] This invention also proposes an event-triggered multi-channel visual inspection system for bridge expansion joints, comprising: The image acquisition module is used to deploy multiple cameras on both sides of the telescopic device, establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. The image unfolding module is used to generate event trigger signals based on the load changes, vibration amplitude or traffic impact conditions of the bridge, and to acquire images of key load conditions. It automatically identifies the four corner points through the key point detection model, sorts the four corner points and constructs a perspective transformation matrix. The perspective transformation matrix is ​​used to unfold the expansion device area into an orthographic view to obtain a geometrically consistent unfolded image of the expansion device area. The calibration module is used to extract multiple profiles at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each profile, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, and average the scaling factors obtained from all profiles to obtain the adaptive scaling calibration value of the unfolded image. The alarm module performs threshold segmentation and connected component analysis on each profile to identify steel beams. The length of blank pixels between adjacent steel beams is defined as the gap width. Only the same gap observed in at least two profiles is retained as a valid gap. The gap width of the valid gap is replaced with millimeter distance according to the scaling factor to generate a gap sequence corresponding to the unfolded image. Multiple safety indicators of the gap sequence are calculated and compared with the corresponding indicator thresholds. Based on the comparison results, a safety alarm is issued.

[0014] Furthermore, the image acquisition module also includes preprocessing operations on real-time continuous frame images. The preprocessing operations include: eliminating image noise and increasing the contrast of real-time continuous frame images by using Gaussian filtering and histogram equalization. The edge features of lane gaps in real-time continuous frame images are extracted using the Canny edge detection algorithm. Optimize edge continuity of real-time consecutive frame images through morphological operations; Hough transform is used to detect straight line edges, thereby improving the accuracy of edge extraction.

[0015] Furthermore, the image unfolding module also includes: generating timed capture signals according to a preset period, acquiring periodic high-definition images, and timestamping and queuing key load condition images and periodic high-definition images from the same camera to form a unified image input sequence.

[0016] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art: 1. Significantly improved real-time performance and scalability: Through the architecture design of multi-threaded streaming and independent analysis threads, the system can support multi-channel concurrent online monitoring, with four channels monitored simultaneously by default, and can be horizontally expanded to more channels according to actual needs; each monitoring channel has an independent detection cycle and threshold setting, realizing flexible matching between high-frequency monitoring of high-risk sections and low-frequency inspection of general sections; adopting three-section consistency verification and "manual / AI dual-channel corner point" mechanism, the robustness of detection in complex environments such as obstruction, reflection and stains is improved, ensuring the reliability of monitoring results.

[0017] 2. High measurement accuracy and high degree of automation: Calibration is performed using the existing dimensions of the structure, eliminating the need for additional visual targets. Combined with Otsu segmentation and connected component analysis methods, fully automatic scale estimation is achieved, avoiding construction interference and error accumulation caused by manual application of scales. By sampling at three locations and averaging, the impact of local dirt and uneven lighting on measurement accuracy can be effectively reduced, ensuring that long-term operating errors are controllable, and the gap measurement deviation is controlled within 0.5 mm.

[0018] 3. Strong data analysis capabilities, supporting early warning and maintenance decisions: The collected data can be subjected to multidimensional statistical analysis. In addition to the conventional maximum, minimum, and average values, range and variance indices are introduced to evaluate the dispersion and uniformity of the joints. By observing the trend of uniformity dispersion over time, potential failures and blockage risks in the expansion joint displacement control system can be identified. Combined with threshold strategies, real-time early warnings of exceeding limits can be achieved, providing a basis for early intervention such as watertight strip damage, pull-out, and local steel component deformation. Historical data can be exported as CSV files for further modeling and analysis, such as eliminating the impact of seasonal temperature differences and retrospectively analyzing traffic impact events, thereby providing reliable data support for the formulation of maintenance plans. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a system structure diagram of Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of image acquisition and processing according to the present invention; Figure 4 This is a schematic diagram of the camera layout of the present invention; Figure 5 This is a schematic diagram of image acquisition according to the present invention; Figure 6 This is a schematic diagram of the geometric dimension feature extraction of the present invention; Figure 7 This is a schematic diagram illustrating the calculation of the gap size in this invention; Figure 8 This is a schematic diagram illustrating the statistical trend of gaps in this invention; Figure 9 This is a schematic diagram of the data management and alarm system of the present invention; Figure 10 This is a schematic diagram of the alarm prompt of the present invention; Figure 11 This is a schematic diagram of the parameter settings for pulling RTSP video streams according to the present invention. Detailed Implementation

[0020] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0021] The method provided by this invention can be implemented in a terminal environment that may include one or more of the following components: a processor, a storage medium, and a display screen. The storage medium stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.

[0022] A processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the terminal, and performs various functions and processes data by running or executing instructions, programs, code sets or instruction sets stored in the storage medium, and by calling data stored in the storage medium.

[0023] Storage media can include random access memory (RAM) or read-only memory (ROM). Storage media can be used to store instructions, programs, code, code sets, or instructions.

[0024] The display screen is used to show the user interface of each application.

[0025] In addition, those skilled in the art will understand that the structure of the terminal described above does not constitute a limitation on the terminal. The terminal may include more or fewer components, or combine certain components, or have different component arrangements. For example, the terminal may also include radio frequency circuits, input units, sensors, audio circuits, power supplies, and other components, which will not be described in detail here.

[0026] Glossary: ​​PLM stands for Product Lifecycle Management, a software system used by enterprises to manage information about the entire process of a product from its inception to its retirement.

[0027] Example 1 like Figure 1 As shown, this embodiment proposes a multi-channel visual inspection method for bridge expansion joints based on event triggering, including: Step 101: Install multiple cameras on both sides of the telescopic device (two cameras are installed at the guardrails of the fast lane and the emergency lane respectively. These cameras are high-resolution cameras to ensure that clear images can be obtained under different lighting conditions), establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. Preferably, a multi-channel networked video input interface is established, and real-time continuous frame images are received from each camera through parallel image acquisition threads, including: continuous streaming of RTSP video streams, with capabilities such as disconnection reconnection and buffer control, and each camera running an independent thread to ensure real-time performance and isolation.

[0028] Specifically, step 101 also includes preprocessing the real-time continuous frame image. The preprocessing operation includes: eliminating image noise and increasing the contrast of the real-time continuous frame image by using Gaussian filtering and histogram equalization. The edge features of lane gaps in real-time continuous frame images are extracted using the Canny edge detection algorithm. Optimize edge continuity of real-time consecutive frame images through morphological operations; Hough transform is used to detect straight line edges, thereby improving the accuracy of edge extraction.

[0029] Step 102: Generate event trigger signals based on the load changes, vibration amplitude, or traffic impact conditions of the bridge, and collect key load condition images. Automatically identify the four corner points through the key point detection model, sort the four corner points and construct a perspective transformation matrix. Expand the expansion joint area into an orthographic view through the perspective transformation matrix to obtain a geometrically consistent expanded image of the expansion joint area. Preferably, the key point detection model is YOLO (Real-time Target Detection Model), which automatically obtains corner points through YOLO key point inference to achieve corner point detection and calibration.

[0030] Alternatively, in this embodiment, the four corner points can also be determined by manual annotation.

[0031] Preferably, the corner points are geometrically sorted, and the perspective transformations cv2.getPerspectiveTransform (used to construct the perspective matrix) and cv2.warpPerspective (both functions in OpenCV) are used to flatten the tilted / perspective distortion expansion joint area into a front view.

[0032] Specifically, step 102 also includes: generating a timed capture signal according to a preset period, acquiring periodic high-definition images, and timestamping and queuing key load condition images and periodic high-definition images from the same camera to form a unified image input sequence.

[0033] Preferably, this embodiment also provides another way to implement the key point detection model, as shown below: 1. Normalize the key load condition images or periodic high-definition images in the image input sequence to obtain the normalized pixel value at each location; 2. A layered attention encoder is set up. This encoder adopts a three-layer structure, with each layer extracting the edge features of the expansion joint's side beams and central beams. The three-layer structure includes: a fine-scale structure layer, a medium-scale structure layer, and a coarse-scale structure layer. The fine-scale structure layer is used to reinforce the small vertical edges of the expansion joint (joint edge, beam side), the medium-scale structure layer focuses on both the edges and the block structure of the beam, and the coarse-scale structure layer emphasizes the overall longitudinal alignment, reduces noise, and smooths the background, as shown below: Calculate the first Key load condition images or periodic high-resolution images of the layer at location Local response value at , in, For the first Key load condition images or periodic high-resolution images of the layer at location Local response value at that location, The x-axis is... The vertical axis is , For the first Layer convolution kernel at offset The weight of the position, For horizontal offset, For vertical offset, For key load condition images or periodic high-resolution images at location The normalized pixel value.

[0034] The convolution kernel is a weighted matrix defined on a local neighborhood, where each element... This represents the contribution weight of neighboring pixels to the response at the current position, where , This represents the dimensionless offset in the horizontal and vertical directions. The convolution kernel obtains the local response value by weighted summation of neighboring pixels. It is used to extract image structural features in a specific direction or scale.

[0035] No. Layer convolution kernel at offset Weight at point Specifically, it includes: , in, For the first Layer weights For the first Lateral offset of the layer The attenuation coefficient, For the first Longitudinal offset of the layer The attenuation coefficient, For the first Layer in offset The structural symbol matrix at the location is used to highlight the strip-shaped structure of the expansion joint.

[0036] If it is a fine-scale structural layer, then for: , It can be 1.0. It can be 0.5. It can be 0.2; If it is a mesoscale structural layer, then for: , It can be 0.8. It can be 0.35. It can be 0.35; If it is a coarse-scale structural layer, then for: , It can be 0.6. It can be 0.2. It can be 0.6; For each layer Perform a weighted summation and use the summation result as the fused response value.

[0037] 3. Configure the geometric constraint decoder, which is used to convert the fused response values ​​into coarse localization of the four corner points. Specifically, this includes: 3.1 Constructing a location confidence plot: , in, For position Confidence level of a pixel. For the Sigmoid function, This is the scaling factor (example value can be 5).

[0038] 3.2 Candidate Corner Generation: Location confidence Figure 2 Value-based transformation yields a binary image: , in, For position A binary mask for each pixel, used to determine its position. Does it belong to the candidate region where corners may exist? This is the confidence threshold.

[0039] Find all Connected regions with a value of 1 (for example, a connected region can be a four-neighbor region (up, down, left, right) or an eight-neighbor region (up, down, left, right + diagonal)) are considered as candidate regions. , This represents the number of candidate regions.

[0040] Calculate the centroid of each candidate region. Each candidate point serves as a candidate point for each candidate region, and the candidate points from all candidate regions form a candidate point set. ,in, For the first The centroid of each candidate region For the first The x-coordinate of the centroid of each candidate region. For the first The ordinate of the centroid of each candidate region.

[0041] 4. Construct a four-point consistent inference engine: Candidate point set The mapping is represented as a set of four corner points, and the optimal combination of four points is selected through geometric consistency scoring, specifically as follows: For any four points in the candidate point set Calculate the geometric consistency score: , in, For geometric consistency score, The weights of the rectangularity index, Rectangularity index The weights for the sequential consistency index, For sequential consistency, The weights of the diagonal symmetry index, It is an index of diagonal symmetry.

[0042] Calculate the rectangularity index Specifically: , , in, Candidate points and distance, Candidate points and distance, Candidate points and distance, Candidate points and The distance.

[0043] Calculate the order consistency index Specifically: , Calculate the diagonal symmetry index include: , , in, Candidate points and distance, Candidate points and The distance.

[0044] 5. Optimal corner selection: Find the maximum geometric consistency score among all candidate four-point combinations; the corresponding candidate four-point combinations are the four final corner points. , in, As the first candidate point, As the second candidate point, As the third candidate point, It is the fourth candidate point.

[0045] 6. Output and Coordinate Mapping Output the set of the four final corner points Each point is a normalized coordinate. This is then back-mapped back to the dimensions of the key load condition image or periodic high-resolution image: , in, For the first The x-coordinates of the corner points after mapping For the first The normalized x-coordinates of the corner points The width of the key load condition image or periodic high-definition image. For the first The ordinates of the corner points after mapping For the first Normalized ordinates of the corner points The height of key load condition images or periodic high-definition images.

[0046] Step 103: Extract multiple cross-sections at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each cross-section, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, average the scaling factors obtained from all cross-sections, and obtain the adaptive scaling calibration value of the unfolded image. Preferably, based on the known nominal width of the steel beam (configurable, default 80mm), the average beam width pixel value is estimated on multiple sampled horizontal lines (25% / 50% / 75% height) using Otsu adaptive thresholding and connected component analysis on the unfolded image. The scaling factor between the average beam width pixel value and the millimeter scale is calculated. The scaling factor obtained from all cross sections is averaged to obtain the adaptive scaling calibration value of the unfolded image, thus completing the targetless self-calibration.

[0047] Step 104: Perform threshold segmentation and connected component analysis on each profile to identify steel beams. Define the blank pixel length between adjacent steel beams as the gap width. Only retain the same gap that is observed in at least two profiles as a valid gap. Replace the gap width of the valid gap with millimeter distance according to the scaling factor to generate the gap sequence corresponding to the unfolded image. Calculate multiple safety indicators for the gap sequence and compare the safety indicators with the corresponding indicator thresholds. Based on the comparison results, issue a safety alarm.

[0048] Specifically, step 104 includes: eliminating false beams and determining valid beams based on the expected beam width range, and defining the blank pixel length between adjacent valid beams as the seam width.

[0049] Preferably, “real beam” (effective beam) sections are selected based on the expected beam width range (e.g., ±50%) (to avoid false signals such as asphalt contamination, water accumulation, and reflection); the length of blank pixels between adjacent real beams is defined as the joint width (pixels); for the same numbered joint, only joints that appear ≥2 times are retained on the three cross sections to improve stability and occlusion robustness.

[0050] Specifically, several safety indicators include: the maximum, minimum, average, range, and variance of the gap sequence.

[0051] Specifically, comparing security indicators with corresponding threshold values ​​includes setting minimum and maximum threshold values ​​for each camera. If any security indicator of a camera is less than the corresponding minimum threshold value or greater than the maximum threshold value, an alarm is triggered, alarm logs are output synchronously, and the alarm is highlighted in the visualization interface.

[0052] Preferably, alarm logs are bidirectionally bridged to the UI text window via _QtTextEditLogHandler and simultaneously printed to the console, facilitating runtime tracking and auditing.

[0053] Specifically, this also includes storing all safety indicators and gap sequences to generate multi-curve trend charts that evolve over time.

[0054] Preferably, it provides a multi-curve trend chart with the real time axis as the horizontal axis, supports switching between "10m / 30m / 1h / 6h / 24h / All" windows, and features such as a monitoring screen, inspection visualization overlay (three profile lines and gaps marked with polygonal patches and their millimeter values), one-click trend viewing, manual corner point interactive annotation, and independent inspection cycle settings for each channel.

[0055] Preferably, all parameters (RTSP, steel beam width, threshold, detection cycle, artificial corner points, etc.) are persisted to config.json, and all cameras can be exported as CSV (time, gap sequence) with one click to meet the requirements of post-evaluation, reporting and regulatory retention.

[0056] Example 2 like Figure 2 As shown, this embodiment proposes an event-triggered multi-channel visual inspection system for bridge expansion joints, including: The image acquisition module is used to deploy multiple cameras on both sides of the telescopic device, establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. The image unfolding module is used to generate event trigger signals based on the load changes, vibration amplitude or traffic impact conditions of the bridge, and to acquire images of key load conditions. It automatically identifies the four corner points through the key point detection model, sorts the four corner points and constructs a perspective transformation matrix. The perspective transformation matrix is ​​used to unfold the expansion device area into an orthographic view to obtain a geometrically consistent unfolded image of the expansion device area. The calibration module is used to extract multiple profiles at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each profile, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, and average the scaling factors obtained from all profiles to obtain the adaptive scaling calibration value of the unfolded image. The alarm module performs threshold segmentation and connected component analysis on each profile to identify steel beams. The length of blank pixels between adjacent steel beams is defined as the gap width. Only the same gap observed in at least two profiles is retained as a valid gap. The gap width of the valid gap is replaced with millimeter distance according to the scaling factor to generate a gap sequence corresponding to the unfolded image. Multiple safety indicators of the gap sequence are calculated and compared with the corresponding indicator thresholds. Based on the comparison results, a safety alarm is issued.

[0057] Other solutions in Example 2 correspond to those in Example 1, so they will not be described again in this example.

[0058] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0059] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0060] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0061] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0062] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, optical disks, and other media capable of storing program code.

[0063] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A multi-channel visual inspection method for bridge expansion joints based on event triggering, characterized in that, include: Step 101: Install multiple cameras on both sides of the telescopic device, establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. Step 102: Generate event trigger signals based on the load changes, vibration amplitude, or traffic impact conditions of the bridge, and collect key load condition images. Automatically identify the four corner points through the key point detection model, sort the four corner points and construct a perspective transformation matrix. Expand the expansion joint area into an orthographic view through the perspective transformation matrix to obtain a geometrically consistent expanded image of the expansion joint area. Step 103: Extract multiple cross-sections at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each cross-section, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, average the scaling factors obtained from all cross-sections, and obtain the adaptive scaling calibration value of the unfolded image. Step 104: Perform threshold segmentation and connected component analysis on each profile to identify steel beams. Define the blank pixel length between adjacent steel beams as the gap width. Only retain the same gap that is observed in at least two profiles as a valid gap. Replace the gap width of the valid gap with millimeter distance according to the scaling factor to generate the gap sequence corresponding to the unfolded image. Calculate multiple safety indicators for the gap sequence and compare the safety indicators with the corresponding indicator thresholds. Based on the comparison results, issue a safety alarm.

2. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, Step 101 also includes preprocessing the real-time continuous frame images. The preprocessing operations include: eliminating image noise and increasing the contrast of the real-time continuous frame images by using Gaussian filtering and histogram equalization. The edge features of lane gaps in real-time continuous frame images are extracted using the Canny edge detection algorithm. Optimize edge continuity of real-time consecutive frame images through morphological operations; Hough transform is used to detect straight line edges, thereby improving the accuracy of edge extraction.

3. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, Step 102 also includes: generating timed capture signals according to a preset period, acquiring periodic high-definition images, and timestamping and queuing key load condition images and periodic high-definition images from the same camera to form a unified image input sequence.

4. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, Step 104 includes: eliminating false beams and determining valid beams based on the expected beam width range, and defining the blank pixel length between adjacent valid beams as the seam width.

5. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, Several safety metrics include: maximum, minimum, average, range, and variance of the gap sequence.

6. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, The comparison of safety indicators with corresponding threshold values ​​includes: setting minimum and maximum threshold values ​​for each camera; if any safety indicator of each camera is less than the corresponding minimum threshold value or greater than the maximum threshold value, an alarm is triggered, alarm logs are output synchronously, and the alarm is highlighted in the visualization interface.

7. The event-triggered multi-channel visual inspection method for bridge expansion joints as described in claim 1, characterized in that, Also includes: All safety indicators and gap sequences are stored to generate a multi-curve trend chart that evolves over time.

8. A multi-channel visual inspection system for bridge expansion joints based on event triggering, characterized in that, include: The image acquisition module is used to deploy multiple cameras on both sides of the telescopic device, establish a multi-channel networked video input interface, and receive real-time continuous frame images from each camera through a parallel image acquisition thread. The image unfolding module is used to generate event trigger signals based on the load changes, vibration amplitude or traffic impact conditions of the bridge, and to acquire images of key load conditions. It automatically identifies the four corner points through the key point detection model, sorts the four corner points and constructs a perspective transformation matrix. The perspective transformation matrix is ​​used to unfold the expansion device area into an orthographic view to obtain a geometrically consistent unfolded image of the expansion device area. The calibration module is used to extract multiple profiles at a preset height position of the unfolded image, perform grayscale conversion, threshold segmentation and connected component analysis on each profile, estimate the average pixel value of the beam width based on the known nominal width of the steel beam, calculate the scaling factor between the average pixel value of the beam width and the millimeter scale, and average the scaling factors obtained from all profiles to obtain the adaptive scaling calibration value of the unfolded image. The alarm module performs threshold segmentation and connected component analysis on each profile to identify steel beams. The length of blank pixels between adjacent steel beams is defined as the gap width. Only the same gap observed in at least two profiles is retained as a valid gap. The gap width of the valid gap is replaced with millimeter distance according to the scaling factor to generate a gap sequence corresponding to the unfolded image. Multiple safety indicators of the gap sequence are calculated and compared with the corresponding indicator thresholds. Based on the comparison results, a safety alarm is issued.

9. A multi-channel visual inspection system for bridge expansion joints based on event triggering as described in claim 8, characterized in that, The image acquisition module also includes preprocessing operations on real-time continuous frame images. The preprocessing operations include: eliminating image noise and increasing the contrast of real-time continuous frame images by using Gaussian filtering and histogram equalization. The edge features of lane gaps in real-time continuous frame images are extracted using the Canny edge detection algorithm. Optimize edge continuity of real-time consecutive frame images through morphological operations; Hough transform is used to detect straight line edges, thereby improving the accuracy of edge extraction.

10. A multi-channel visual inspection method for bridge expansion joints based on event triggering as described in claim 8, characterized in that, The image unfolding module also includes: generating timed capture signals according to a preset period, acquiring periodic high-definition images, and timestamping and queuing key load condition images and periodic high-definition images from the same camera to form a unified image input sequence.

Images