Bridge and monitoring system operation state non-interference traffic detection method and system

Abstract

The invention relates to the field of bridge engineering, and particularly discloses a bridge and a monitoring system operation state non-interference traffic detection method and system. Firstly, test fleet queue parameters are determined according to bridge load test specifications, and dynamic world coordinates of the test fleet running on a bridge surface, videos of the whole process of the test fleet passing through a test bridge and various monitoring data are recorded by the test fleet running through the bridge; then using an algorithm to identify driving tracks of all vehicles on the bridge surface in the bridge crossing process of the test vehicle team, recording camera view track information of the unmanned aerial vehicle, and using a dynamic time warping algorithm to establish a mapping relation; obtaining theoretical response of each monitoring point of the bridge; and actual measurement response components under the traffic flow load effect are obtained and compared, a structure verification coefficient is calculated, and the bridge structure state and the monitoring system precision are synchronously evaluated. According to the method, the detection cost is greatly reduced, the detection efficiency is remarkably improved, and online detection and evaluation can be synchronously carried out on the operation states of the road network bridge group and the monitoring system thereof.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge engineering, and particularly relates to a bridge and a method and system for detecting the operating state of a monitoring system without interfering with traffic. Background Art

[0002] Highway road networks undertake the important function of transporting the public. Bridges are key nodes for crossing obstacles in road networks. Accurately evaluating the state of highway bridges is very important for ensuring the connectivity of road networks and the safety of people's lives and property. The load test is currently the most direct and effective means for bridge state evaluation. A bridge load test refers to evaluating the structural state by directly comparing the calculated mechanical responses at the structural control parts with the monitored actual responses under the input of a deterministic load with known spatial and temporal distributions (reaching a certain load efficiency threshold). It is widely recognized in the current bridge technology field because of its direct and effective method. However, traditional load tests require traffic interruption and static loading of heavy objects with a preset spatial and temporal distribution on the bridge deck. This process consumes a large amount of manpower and material resources, has low efficiency and interferes with traffic, making it difficult to carry out regularly and cover a large number of bridges, resulting in difficulty in timely and effective evaluation of bridge states.

[0003] On the other hand, structural health monitoring technology has gradually emerged as the most common method for implementing monitoring and early warning of bridge states. Structural health monitoring systems are widely used in the health management of bridges. Especially in the past five years in China, the construction of bridge health monitoring systems has been vigorously promoted nationwide. However, during the operation of bridge health monitoring systems, they are often affected by factors such as sensor lifespan, stability, and environmental interference, resulting in a large amount of abnormal noise in the monitoring data. It is very difficult for engineering personnel to distinguish whether the abnormal monitoring data is caused by abnormal structural states or abnormal sensor systems, resulting in many abnormal alarms in the monitoring system, which is very unfavorable for accurately grasping the structural health state. Currently, in order to check whether there are abnormalities in the structural monitoring system, the method of on-site inspection by personnel at the bridge site is usually adopted, which consumes a great deal of human and material costs. Therefore, there is a need to develop a more effective verification method that does not affect the normal operation of the monitoring system.

[0004] To sum up, currently, whether it is the bridge health state or the state of the bridge health monitoring system, time-consuming and laborious technical means such as manual verification are still used. On the one hand, it interferes with normal operating traffic, and on the other hand, due to the time-consuming and laborious nature, it cannot be widely applied to road network bridges, greatly restricting the true and effective service of the health monitoring system for the purpose of bridge health state monitoring and early warning. Summary of the Invention

[0005] In view of the defects existing in the above-mentioned prior art, the present invention proposes a method and system for detecting the operating state of a bridge and its monitoring system without disturbing traffic. Through the accurate spatio-temporal mapping of the vehicle fleet load on the bridge deck and the structural monitoring response, a non-disturbing bridge load test is realized, and the problem of evaluating the operating state of the bridge and its health monitoring system is solved synchronously. It can effectively solve the problem that traditional load tests and the verification of bridge health monitoring systems require traffic closure, and has technical advantages such as high implementation efficiency, low economic cost, no traffic interference, and a wide range of bridges benefited.

[0006] The above object is achieved by the following technical solutions:

[0007] The method and system for detecting the operating state of the bridge and its monitoring system without disturbing traffic according to the present invention include the following technical steps:

[0008] S1. Determine the load test efficiency threshold value according to the bridge load test specification, and set the test vehicle fleet queue parameters according to this threshold value, including the number of trucks, total weight, axle weight, wheelbase, following distance, lane layout, and driving speed;

[0009] S2. Drive the test vehicle fleet across the bridge according to the fleet loading plan. An RTK positioning system is arranged on a leading truck in the test vehicle fleet to record its dynamic world coordinates during driving on the bridge deck. The unmanned aerial vehicle hovers above the bridge to aerial photograph the whole process video of the test vehicle fleet passing through the test bridge, and the health monitoring system records various monitoring data during the process of the test vehicle fleet passing through the bridge;

[0010] S3. Use the YOLOv11 and ByteTrack algorithms to identify the driving trajectories of all vehicles on the bridge deck during the process of the test vehicle fleet passing through the bridge. According to the dynamic world coordinates of the leading truck equipped with the RTK system and its trajectory information in the field of view of the unmanned aerial vehicle camera, establish a mapping relationship using the dynamic time warping algorithm;

[0011] S4. Apply the mapping relationship established in step S3 to the video recognition trajectory information of other vehicles passing through the bridge in the test vehicle fleet to obtain the dynamic world coordinates of all vehicles in the test vehicle fleet during driving on the bridge deck, and load them into the bridge finite element model to obtain the theoretical responses of each monitoring point of the bridge;

[0012] S5. Separate the monitoring data recorded by the health monitoring system in step S2 to obtain the measured response components under the action of the vehicle flow load, perform dynamic time warping with the theoretical response obtained in step S4 to align their sequences, compare the amplitudes, calculate the structural calibration coefficient, and evaluate the bridge structure state and the monitoring system accuracy synchronously.

[0013] Further, the load test efficiency threshold value in step S1 is η min , and the load test efficiency η of the test vehicle fleet is calculated by the formula η = L a / [L d(1 + μ)], where L a is the maximum value of the calculated load effect of the test vehicle fleet crossing the bridge, and L d is the maximum value of the calculated load effect of the vehicle load model in the bridge design code. μ is the vehicle load impact factor calculated in the bridge design code; by adjusting the queue parameters of the test vehicle fleet, the test efficiency η of the test vehicle fleet is made to satisfy η ≥ η min , and the queue parameters of the test vehicle fleet include the number of trucks, total weight, axle weight, wheelbase, following distance, lane arrangement, and driving speed.

[0014] Furthermore, the YOLOv11 algorithm in step S3 is used to identify vehicle contour information through instance segmentation from the perspective of the UAV, specifically including:

[0015] Step 311: Use the UAV to pre-take vehicle images from multiple angles as the training data set, accurately label the vehicle contour area in each image, and record the center point coordinates (x cl , y cl ) and width and height (w l , h l ) of the labeled rectangle box. Randomly divide the labeled data set into a training set, a test set, and a validation set according to 80%, 10%, and 10%;

[0016] Step 312: Train the YOLOv11 network model through the training set. In each batch of training, based on the YOLO model, predict the center point coordinates (x cp , y cp ) and width and height (w p , h p ) of the vehicle area in the image, and calculate the corresponding localization loss classification loss and confidence classification loss comprehensive loss The calculation formulas are as follows:

[0017]

[0018] In the formula, λ coord is the weight coefficient for weighing different losses; S is the number of grids of the feature map extracted by the model; B is the number of bounding boxes predicted by each grid; is to indicate whether the jth bounding box in the ith grid contains the target. If it contains the target, it is 1, otherwise it is 0; x cp,i , y cp,i are the center point coordinates of the bounding box of the vehicle area in the ith grid predicted by the YOLO model for the image; x cl,i , y cl,i are the center point coordinates of the bounding box of the vehicle area in the ith grid of the image annotation; w p,i , h p,iis the width and height of the bounding box of the vehicle area in the i-th grid of the YOLO model's predicted image; w l,i , h l,i is the width and height of the bounding box of the vehicle area in the i-th grid of the image annotation; C represents the label of the annotated vehicle contour, is an indicator of whether the i-th grid contains a target, 1 if it does, otherwise 0, p i (C) is the confidence of the bounding box of label category C in the i-th grid predicted by the model; is the true confidence of the target category in the i-th grid; M is the number of samples in the training set; y mc is the true confidence of the c-th category of the m-th sample; α is the positive and negative sample balance coefficient; γ is the small sample focusing coefficient; classes is the label of the vehicle contour area;

[0019] Calculate the comprehensive loss through backpropagation The gradient of the weight W

[0020]

[0021] Update the model weights of the corresponding training batch through the Adam optimizer:

[0022]

[0023] where ρ is the generalized learning rate, W q+1 is the model weight of the q + 1 batch; W q is the model weight of the q batch;

[0024] Step 313, Record the comprehensive loss of each training batch and the model weight W, and find the weight that minimizes the comprehensive loss through cyclic training, denoted as the optimal weight W optmal , and use the YOLOv11 model configured with the optimal weight W optmal to identify the images in the whole-process video of the drone hovering over the bridge and the test vehicle convoy passing through the test bridge obtained in step S2, and identify the center point coordinates (x ca,n , y ca,n ) and the width and height (w a,n , h a,n ) of the bounding boxes of the contours, wheels, and license plate areas of the n-th vehicle, and complete the instance segmentation of the vehicle contour information.

[0025] Furthermore, the ByteTrack algorithm described in step S3 is used to track the vehicles as targets, and obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle convoy crossing the bridge, specifically including:

[0026] Step 321, Based on the configured optimal weight Woptmal The YOLOv11 model identifies the nth vehicle in the t-th frame image of the test vehicle convoy crossing the bridge, and extracts the center point coordinates (x ca,n,t , y ca,n,t ) of its contour bounding box as the trajectory point;

[0027] Step 322: Input the center point coordinates (x ca,n,t , y ca,n,t ) of the contour bounding box of the nth vehicle in the t-th frame image into the ByteTrack object tracking algorithm, and use the Kalman filter to predict the center point coordinates (x ck,n,t , y ck,n,t ) of the vehicle in the t-th frame:

[0028]

[0029] where F t is the state transition matrix; B t is the control input matrix; v t is the control vector; P t|t-1 is the prediction covariance matrix; Q t is the process noise covariance matrix; (x cf,n,t-1 , y cf,n,t-1 ) is the corrected trajectory state updated by the Kalman filter in the (t - 1)-th frame, and the superscript T represents the transpose of the matrix, and P t-1 represents the prediction covariance matrix in the (t - 1)-th frame;

[0030] Step 323: Calculate the intersection over union (IoU) of the bounding box (x ca,n,t ±w 2,n , y ca,n,t ±h a,n ) and the predicted bounding box (x ck,n,t ±w a,n , y ck,n,t ±h a,n ), which is the ratio of the area of intersection (AoI) to the area of union (AoU) of the bounding boxes:

[0031]

[0032] where AoI1 and AoI2 represent the overlapping lengths of the two bounding boxes in the horizontal and vertical directions respectively; h a,n represents the width and height of the contour bounding box of the nth vehicle;

[0033] Step 324: For the detection box with an IoU value exceeding the set threshold IoU limit , update the corrected trajectory state (x cf,n,t , y cf,n,t ) of the nth vehicle in the t-th frame through the Kalman filter:

[0034]

[0035] In the formula, K t is the Kalman gain matrix; H t is the observation matrix; R t is the observation noise covariance matrix, and I represents the identity matrix;

[0036] Step 325: Delete the vehicle trajectories that have not been matched for more than 5 seconds, and output the matching tracking results of the nth vehicle, including the vehicle ID, vehicle speed (calculated by the difference in the center point coordinates of the front and rear frames), and the vehicle trajectory {(x cf,n,t , y cf,n,t ) | y = 0, 1, 2,...}; Loop through steps 321 to 325 to obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle fleet crossing the bridge.

[0037] Furthermore, the dynamic time warping algorithm described in step S3 is based on the dynamic world coordinates of the leading truck installed with the RTK system, and establishes a conversion mapping relationship from the trajectory coordinates of the leading truck in the UAV captured image to the world coordinates recorded by the RTK. This conversion relationship is the optimal warping path S that enables the two time coordinate sequences to match in the dynamic time warping algorithm load , specifically including:

[0038] Step 331: Take the world coordinates recorded by the leading truck installed with the RTK system as the leading truck RTK vehicle trajectory {(x crtk,l,t , y crtk,l,t ) | t = 0, 1, 2,...}, and through resampling, make the time series lengths of the leading truck in the UAV captured image, the YOLOv1 1 model, and the leading vehicle trajectory {(x cf,l,t , y cf,l,t ) | t = 0, 1, 2,...} identified by the ByteTrack algorithm consistent, and construct the Euclidean distance two-dimensional matrix D crtk,l,t , y crtk,l,t ) | t = 0, 1, 2,...} and {(x cf,l,t , y cf,l,t ) | t = 0, 1, 2,...}: L×L :

[0039]

[0040] In the formula, u is the index of the RTK trajectory sequence; g is the index of the UAV trajectory sequence;

[0041] Step 332: Recursively construct the cumulative distance matrix C L×L based on the matrix D L×L :

[0042] C L×L (u, g) = DL×L (u,g) + min{C(u - 1,g), C(u,g - 1), C(u - 1,g - 1)}

[0043] Find the cumulative distance matrix C L×L The path with the minimum value from the upper left corner to the lower right corner (the shortest path) is the optimal warping path S for the matching of RTK and UAV trajectory sequences load 。

[0044] Furthermore, applying the mapping relationship established in step S3 to the video recognition trajectory information of other vehicles crossing the bridge in the test convoy specifically includes:

[0045] Step 411: Based on the spatio-temporal trajectory information of all vehicles on the bridge deck during the entire process of the test convoy crossing the bridge, construct the vehicle recognition trajectory x-coordinate matrix X video,Z×K and the y-coordinate matrix Y video,Z×K :

[0046]

[0047] In the matrix, K is the total number of vehicle flows on the bridge deck, where k = 1 is the leading vehicle; Z is the time span from the leading vehicle getting on the bridge to the last vehicle of the convoy getting off the bridge; and respectively represent the trajectory x-coordinate and trajectory y-coordinate of the k-th vehicle at time z;

[0048] Step 412: Based on the obtained optimal warping path S load , synchronously adjust the sequence repetition items in the x and y coordinate matrices of the vehicle recognition trajectory, and adjust the number of rows of the matrix to realize the coordinate regularization of the overall vehicle flow on the bridge deck, and obtain the x and y world coordinate matrices X real,Z×K , Y real,Z×K :

[0049]

[0050] Furthermore, obtaining the dynamic world coordinates of all vehicles in the test convoy driving on the bridge deck and loading them into the bridge finite element model to obtain the theoretical responses of each monitoring point of the bridge specifically includes:

[0051] Step 421: Based on the test convoy queue parameters formulated in step S1, combined with the x world coordinate matrix X real,Z×K and the y world coordinate matrix Y real,Z×K of all vehicles crossing the bridge obtained in step 72, construct the multi-source vehicle flow load dynamic time series S(z):

[0052]

[0053] In the formula, are the x and y world coordinates of the k-th vehicle at time z; b k is the number of axles of the k-th vehicle; is the distance between the β-th axle and the track x coordinate and the track y coordinate at time z; is the axle weight load of the β-th axle;

[0054] Step 422: Extract the influence surface functions I of each monitoring point of the bridge, i.e., displacement, deflection, and strain measurement points m =(x b , y b ), where m is the monitoring point number, and I m =(x b , y b ) represents the response value of the measurement point m when a unit concentrated load is applied at the world coordinates (x b , y b ) of the bridge deck; Calculate the theoretical response time series R m (z) based on the multivariate vehicle flow load dynamic time series constructed in Step 81:

[0055]

[0056] In the formula, r z is the calculated theoretical response value of the measurement point m at time z, including displacement, deflection, and strain; is the x-direction coordinate of the β-th axle of the k-th vehicle on the bridge deck at time z; is the y-direction coordinate of the i-th axle of the k-th vehicle on the bridge deck at time t;

[0057] Furthermore, the signal separation of the monitoring data recorded by the health monitoring system described in Step S5 to obtain the measured response component under the action of the vehicle flow load and the dynamic time warping with the theoretical response obtained in Step S4 to align their sequences specifically includes:

[0058] Step 51: Resample the monitoring data vehicle flow load response time series M m ={m z |z = 0, 1, 2,...} of the measurement point m to be the same length as the calculated theoretical response time series R m ={r z |z = 0, 1, 2,...} of the measurement point m, and construct the Euclidean distance matrix O m of M m and R Z×Z :

[0059]

[0060] In the formula, is the index for monitoring the response sequence; ∈ is the index for calculating the theoretical response sequence;

[0061] Step 52, based on the matrix O Z×Z Recursively construct the cumulative distance matrix V Z×Z :

[0062]

[0063] Find the cumulative distance matrix V Z×Z The path with the minimum value from the upper left corner to the lower right corner (the shortest path) of the cumulative distance matrix V is the optimal warping path A between the monitored response sequence and the calculated theoretical response sequence load ; Delete sequence duplicates and adjust R m The sequence length of to achieve fine-tuning matching between response time series;

[0064] The calculation formula for the structure check coefficient ψ of Step S5 is ψ = (ξ t - ξ r ) / ξ c , where ξ t is the maximum value of the quasi-static response M of the traffic flow load under the action of the open traffic moving load of the test vehicle fleet m , ξ r is the average value of the quasi-static response of the traffic flow load when there is no vehicle or only a car on the bridge deck after the test vehicle fleet leaves the bridge deck, and ξ c is the maximum value of the theoretical calculated load effect R of the corresponding test vehicle fleet and environmental vehicles m .

[0065] The present invention also provides a bridge and its monitoring system for detecting the operating state without interfering with traffic. This system is loaded with the above method. This system includes four components: an artificial test vehicle fleet subsystem, a vehicle spatio-temporal trajectory acquisition subsystem, a vehicle trajectory coordinate mapping subsystem, and a bridge and its monitoring system evaluation subsystem. The artificial test vehicle fleet subsystem is the carrier of the detection, which is composed of several trucks of the same model according to the designed loading scheme. The vehicle spatio-temporal trajectory acquisition subsystem obtains the driving trajectory of the artificial test vehicle fleet under the vision of the unmanned aerial vehicle and the dynamic world coordinates recorded by the RTK of the leading vehicle. The vehicle trajectory coordinate mapping subsystem uses the dynamic time warping algorithm to calculate the dynamic world coordinates of all vehicles crossing the bridge based on the information obtained by the position trajectory acquisition subsystem, and loads the dynamic world coordinates of all vehicles crossing the bridge into the bridge finite element model to obtain the theoretical traffic flow load response of the monitoring points. The bridge and its monitoring system evaluation subsystem calculates the structure check coefficient through the dynamic warping theory and the measured traffic flow load response sequence, and determines the operating state of the bridge and its monitoring system according to the calculated value of the structure check coefficient.

[0066] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for detecting the operating state of a highway bridge and its health monitoring system without disturbing traffic, having the following beneficial effects:

[0067] (1) The present invention organizes and utilizes a test vehicle fleet to conduct load test detection and evaluation on the bridge and its health monitoring system under the condition of not disturbing traffic, solving the technical defects of traditional test detection methods such as traffic interruption, high human and material costs, low implementation frequency, and small coverage of bridges.

[0068] (2) The load test of the test vehicle fleet in the present invention can economically and efficiently cover all bridge structures of the expressway network and ordinary highway network, having advantages such as high universality, simple and clear method, and low economic cost. It can provide an innovative technical means for carrying out test detection and evaluation of bridges and their health monitoring systems for a large number of road network bridge clusters under the condition of not disturbing traffic, and has a broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flowchart of a method for detecting the operating state of a highway bridge and its health monitoring system without disturbing traffic;

[0070] Figure 2 It is an arrangement diagram of a case bridge and its health monitoring system;

[0071] Figure 3 It is a schematic diagram of the fleet arrangement and vehicle load of the artificial test vehicle fleet subsystem;

[0072] Figure 4 It is a live-action picture of the test vehicle fleet crossing the bridge taken by a drone;

[0073] Figure 5 It is a dynamic world coordinate trajectory diagram of the leading vehicle of the fleet recorded by an RTK device;

[0074] Figure 6 It is a schematic diagram of the instance segmentation and recognition of the vehicle contour on the bridge deck by the YOLOv11 machine vision algorithm of the present invention;

[0075] Figure 7 It is a flowchart of the dynamic time programming algorithm of the present invention for regularizing the video recognition trajectory and the RTK recorded trajectory;

[0076] Figure 8 It is a schematic diagram of the spatio-temporal distribution of the fleet calibrated by the dynamic time programming of the present invention;

[0077] Figure 9 It is a schematic diagram of the separation of the measured load effect of the bridge and its time history signal (taking the deflection measurement point DR#7 as an example);

[0078] Figure 10Flow chart of the dynamic time programming algorithm's regularized theory response and measured response of the present invention;

[0079] Figure 11 Schematic diagram of the theoretical and measured response curves and related parameters of the present invention after dynamic time warping;

[0080] Figure 12 Structural diagram of a highway bridge and its health monitoring system's operation state non-interference traffic detection system. Detailed implementation manners

[0081] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0082] Embodiment 1:

[0083] As Figure 1 shown, the method for detecting the operation state of the bridge and its monitoring system without interfering with traffic provided in this embodiment includes the following steps:

[0084] Step S1: Determine the load test efficiency threshold value according to the bridge load test specification, and set the test vehicle fleet loading plan according to this threshold value; Arrangement of the technical condition of the bridge in the embodiment and its health monitoring system Figure 2 shown, the bridge is a separated concrete continuous rigid frame box girder bridge, with a total length of 374m and a total width of 33.2m. It adopts a one-way three-lane and emergency parking lane design, and is equipped with a structural deflection and strain monitoring system;

[0085] Establish a structural finite element model according to the bridge design drawings, and the load test efficiency threshold value is η min , and the load test efficiency η of the test vehicle fleet is calculated by the formula η = L a / [L d (1 + μ)], where L a is the maximum value of the calculated load effect of the test vehicle fleet crossing the bridge, L d is the maximum value of the calculated load effect of the vehicle load model in the bridge design specification, and μ is the vehicle load impact coefficient calculated by the bridge design specification; By adjusting the queue parameters of the test vehicle fleet, make the load test efficiency η of the test vehicle fleet satisfy η ≥ η min , and the queue parameters of the test vehicle fleet include the number of trucks, total weight, axle weight, wheelbase, following distance, lane arrangement, driving speed. At least one lane should be opened for environmental vehicles. The fleet should keep moving across the bridge evenly with a speed ≤ 30 km / h, and the following distance is recommended to be 10 - 30m and ensure driving safety. The total vehicle weight and axle weight should take the maximum value under the specification requirements of not overloading; Finally, the queue parameters of the test vehicle fleet for the embodiment bridge are as Figure 3 shown, and the load test efficiency is as shown in the following table:

[0086] <![CDATA[Load test efficiency threshold value η min > 0.7 Mid-span load test efficiency (displacement) 0.80 Mid-span load test efficiency (bending moment) 0.78 Side-span mid-span load test efficiency (displacement) 0.69 Side-span mid-span load test efficiency (bending moment) 0.72

[0087] As can be seen from the table, the load test efficiency of the key sections of the bridge in the embodiments under the load of the finally selected test vehicle fleet all reaches the load test efficiency threshold value;

[0088] Step S2: Drive the artificial test vehicle fleet across the bridge according to the loading scheme. The RTK positioning system records the dynamic world coordinates of a leading truck, the drone hovers over the bridge to aerial photograph the whole process video of the test vehicle fleet crossing the bridge, and the health monitoring system records various monitoring data; Figure 4 Shows the actual situation of the test vehicle fleet of the bridge in the embodiment photographed by the drone, and completely records the complete video frame information of the test vehicle fleet entering and leaving the bridge; The dynamic world coordinates recorded by the RTK positioning system of the leading vehicle are as Figure 5 shown. The RTK device is arranged on the roof of the leading vehicle. Its acquisition system can record the dynamic world coordinate information of the leading vehicle. All the acquired data is time-synchronized and covers the whole process of the test vehicle fleet driving on the bridge deck.

[0089] Step S3: Use the machine vision algorithms YOLOv11 and ByteTrack to identify the driving trajectories of all vehicles on the bridge deck, Figure 6 Is the schematic diagram of the instance segmentation and recognition of the vehicle contour on the bridge deck by the YOLOv11 machine vision algorithm of the present invention. It can be seen that the contours of the vehicles crossing the bridge are accurately recorded; The spatio-temporal trajectory information of all vehicles visually recognized by the algorithm is output; The dynamic time warping algorithm is used to establish the mapping relationship between the dynamic world coordinates of the leading vehicle RTK system and the aerial photography trajectory of the drone; The mapping relationship is used for the video recognition trajectory information of other vehicles crossing the bridge to obtain the dynamic world coordinates of all vehicles driving on the bridge deck. The steps are as Figure 7 shown.

[0090] The YOLOv11 algorithm is used to identify the vehicle contour information through instance segmentation from the perspective of the drone, specifically including:

[0091] Step 311: Use the drone to pre-take vehicle images from multiple angles as the training data set, accurately label the vehicle contour areas in each image, and record the center point coordinates (x cl , y cl ) and width and height (w l , h l ) of the labeled rectangular box, and randomly divide the labeled data set into a training set, a test set and a validation set according to 80%, 10% and 10%;

[0092] Step 312: Train the YOLOv11 network model through the training set. In each batch of training, based on the YOLO model, predict the center point coordinates (x cp , y cp ) and width and height (w p,h p ), calculate the corresponding positioning loss Classification loss and confidence classification loss Comprehensive loss The calculation formula is as follows:

[0093]

[0094] In the formula, λ coord is the weight coefficient for weighing different losses; S is the number of grids of the feature map extracted by the model; B is the number of bounding boxes predicted for each grid; indicates whether the j-th bounding box in the i-th grid contains the target, which is 1 if it does, otherwise 0; x cp,i ,y cp,i are the coordinates of the center point of the bounding box of the vehicle area in the i-th grid of the image predicted by the YOLO model; x cl,i ,y c1,i are the coordinates of the center point of the bounding box of the vehicle area in the i-th grid of the image annotation; w p,i ,h p,i are the width and height of the bounding box of the vehicle area in the i-th grid of the image predicted by the YOLO model; w l,i ,h l,i are the width and height of the bounding box of the vehicle area in the i-th grid of the image annotation; C represents the label of the vehicle contour; indicates whether the i-th grid contains the target, which is 1 if it does, otherwise 0, p i (C) is the confidence of the bounding box of label category C in the i-th grid predicted by the model; is the true confidence of the target category in the i-th grid; M is the number of samples in the training set; y mc is the true confidence of the m-th sample in the c-th category; α is the positive and negative sample balance coefficient; γ is the small sample focusing coefficient; calsses is the label of the vehicle contour area;

[0095] Calculate the comprehensive loss through backpropagation The gradient of the weight W

[0096]

[0097] Update the model weights of the corresponding training batch through the Adam optimizer:

[0098]

[0099] In the formula, ρ is the generalized learning rate, W q+1 is the model weight of the q+1 batch; W q is the model weight of the q batch;

[0100] Step 313: Record the comprehensive loss of each training batch and the model weights W. Through cyclic training, find the weights that minimize the comprehensive loss , and denote the minimized weights as the optimal weights W optmal . Use the YOLOv11 model configured with the optimal weights W optmal to identify the images in the video of the entire process of the drone hovering over the bridge and the test vehicle convoy passing through the test bridge obtained in step S2. Identify the center point coordinates (x ca,n , y ca,n ) and the width and height (w a,n , h a,n ) of the bounding box of the contour, wheels, and license plate area of the nth vehicle, and complete the instance segmentation of the vehicle contour information.

[0101] The ByteTrack algorithm is used to perform target tracking on the vehicles to obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the entire process of the test vehicle convoy crossing the bridge, specifically including:

[0102] Step 321: Based on the YOLOv11 model configured with the optimal weights W optmal , identify the nth vehicle in the tth frame image of the test vehicle convoy crossing the bridge video, and extract the center point coordinates (x ca,n,t , y ca,n,t ) of its contour bounding box as the trajectory point;

[0103] Step 322: Input the center point coordinates (x ca,n,t , y ca,n,t ) of the contour bounding box of the nth vehicle in the tth frame image into the ByteTrack target tracking algorithm, and use the Kalman filter to predict the center point coordinates (x ck,n,t , y ck,n,t ) of the vehicle in the tth frame:

[0104]

[0105] In the formula, F t is the state transition matrix; B t is the control input matrix; v t is the control vector; P t|t-1 is the prediction covariance matrix; Q t is the process noise covariance matrix; (x cf,n,t-1 , y cf,n,t-1 ) is the corrected trajectory state updated by the Kalman filter in the (t - 1)th frame, and the superscript T represents the transpose of the matrix, and P t-1 represents the prediction covariance matrix in the (t - 1)th frame;

[0106] Step 323: Calculate the bounding box (x ca,n,t ±w a,n,y ca,n,t ±h a,n ) and the predicted bounding box (x ck,n,t ±w a,n ,y ck,n,t ±h a,n ) is the intersection over union IoU, which is the ratio of the intersection area AoI to the union area AoU of the bounding boxes:

[0107]

[0108] In the formula, AoI1 and AoI2 respectively represent the overlapping lengths of the two bounding boxes in the horizontal and vertical directions; h a,n represents the width and height of the contour bounding box of the nth vehicle;

[0109] Step 324. For the detection boxes with IoU values exceeding the set threshold IoU limit , update the corrected trajectory state (x cf,n,t ,y cf,n,t ) of the nth vehicle in the tth frame through Kalman filtering:

[0110]

[0111] In the formula, K t is the Kalman gain matrix; H t is the observation matrix; R t is the observation noise covariance matrix, and I represents the identity matrix;

[0112] Step 325. Delete the vehicle trajectories that have not been matched for more than 5 seconds, and output the matching tracking results of the nth vehicle, including the vehicle ID, vehicle speed (calculated by the difference between the center point coordinates of the previous and subsequent frames), and vehicle trajectory {(x cf,n,t ,y cf,n,t )|t = 0,1,2,...}; Loop through steps 321 to 325 to obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle convoy crossing the bridge.

[0113] The described dynamic time warping algorithm is based on the dynamic world coordinates of the leading truck installed with the RTK system, and establishes a conversion mapping relationship from the trajectory coordinates of the leading truck's UAV captured image to the RTK-recorded world coordinates. This conversion relationship is the optimal warping path S that enables the two time coordinate sequences to match in the dynamic time warping algorithm load , specifically including:

[0114] Step 331. Use the world coordinates recorded by the leading truck installed with the RTK system as the leading truck RTK vehicle trajectory {(x crtk,l,t ,y crtk,l,t)|t = 0, 1, 2,...}, resample to make the time series lengths of the leading truck drone captured images, the leading vehicle trajectories identified by the YOLOv1 1 model and the ByteTrack algorithm {(x cf,l,t , y cf,l,t )|t = 0, 1, 2,...} consistent, and construct the Euclidean distance two-dimensional matrix D of {(x crtk,l,t , y crtk,l,t )|t = 0, 1, 2,...} and {(x cf,l,t , y cf,l,t )|t = 0, 1, 2,...}: L×L :

[0115]

[0116] In the formula, u is the index of the RTK trajectory sequence; g is the index of the drone trajectory sequence;

[0117] Step 332, based on the matrix D L×L Recursively construct the cumulative distance matrix C L×L :

[0118] C L×L (u, g) = D L×L (u, g) + min{C(u - 1, g), C(u, g - 1), C(u - 1, g - 1)}

[0119] Find the path with the smallest value from the upper left corner to the lower right corner of the cumulative distance matrix C (the shortest path), which is the optimal warping path S for matching the RTK and drone trajectory sequences L×L . load .

[0120] The final spatio-temporal distribution of the calibrated vehicle fleet and the vehicles crossing the bridge by dynamic time warping is as Figure 8 shown.

[0121] Step S4, apply the mapping relationship established in Step S3 to the video recognition trajectory information of other vehicles crossing the bridge in the test vehicle fleet, specifically including:

[0122] Step 411, based on the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle fleet crossing the bridge, construct the vehicle recognition trajectory x coordinate matrix X video,Z×K and the y coordinate matrix Y video,Z×K :

[0123]

[0124] In the matrix, K is the total number of vehicle flows on the bridge deck, where k = 1 is the leading vehicle; Z is the time span from the leading vehicle getting on the bridge to the last vehicle of the fleet getting off the bridge; and respectively represent the trajectory x coordinate and trajectory y coordinate of the k-th vehicle at time z;

[0125] Step 412. Based on the obtained optimal warping path S load , synchronously adjust the sequence duplicates in the x and y coordinate matrices of the vehicle recognition trajectory, and adjust the number of rows of the matrix to achieve the overall coordinate regularization of the bridge deck traffic flow, so as to obtain the x and y world coordinate matrices X real,Z×K , Y real,Z×K :

[0126]

[0127] Obtain the dynamic world coordinates of all vehicles in the test vehicle fleet driving on the bridge deck, and load them into the bridge finite element model to obtain the theoretical responses of each monitoring point of the bridge; specifically including:

[0128] Specifically including:

[0129] Step 421. Based on the test vehicle fleet queue parameters formulated in step S1, combined with the x world coordinate matrix X real,Z×K and y world coordinate matrix Y real,Z×K of all vehicles crossing the bridge obtained in step 72, construct a multivariate traffic flow load dynamic time series S(z):

[0130]

[0131] In the formula, are the x and y world coordinates of the k-th vehicle at time z; b k is the number of axles of the k-th vehicle; is the distance of the β-th axle relative to the trajectory x coordinate and trajectory y coordinate at time z; is the axle weight load of the β-th axle;

[0132] Step 422. Extract the influence surface functions I m =(x b , y b ) of each monitoring point of the bridge, that is, displacement, deflection, and strain measurement points, where m is the monitoring point number, and I m =(x b , y b ) represents the response value of the measurement point m when a unit concentrated load is applied at the bridge deck world coordinates (x b , y b ); based on the multivariate traffic flow load dynamic time series constructed in step 81, calculate the theoretical response time series R m (z):

[0133]

[0134] In the formula, rz is the calculated theoretical response value of measurement point m at time z, including displacement, deflection, and strain; is the x-direction coordinate of the bridge deck at time z of the β-axis of the k-th vehicle; is the y-direction coordinate of the bridge deck at time t of the i-axis of the k-th vehicle.

[0135] Step S5: Separate the monitoring data to obtain the measured response components under the action of vehicle flow load. The measured load effect and its time history separation signal of the deflection measurement point DR#7 of the bridge in the embodiment are as Figure 9 shown, including the quasi-static action of vehicle flow load, low-frequency long-term components (caused by temperature, etc.), and high-frequency vibration components (caused by vehicle coupling vibration, etc.); Step 51: Through resampling, the response time series M of the vehicle flow load acting on the monitoring data of measurement point m m ={m z |z = 0, 1, 2,...} is made to be the same length as the calculated theoretical response time series R m ={r z |z = 0, 1, 2,...}, and the Euclidean distance matrix O m of M m and R Z×Z is constructed:

[0136]

[0137] In the formula, is the index of the monitoring response sequence; ∈ is the index of the calculated theoretical response sequence;

[0138] Step 52: Recursively construct the cumulative distance matrix V Z×Z based on the matrix O Z×Z :

[0139]

[0140] Find the path with the smallest value from the upper left corner to the lower right corner of the cumulative distance matrix V Z×Z (the shortest path), which is the optimal warping path A between the monitoring response sequence and the calculated theoretical response sequence load ; Delete the duplicate items in the sequence and adjust the sequence length of R m to achieve fine-tuning matching between the response time series;

[0141] The structural check coefficient ψ calculation formula in Step S5 is ψ = (ξ t - ξ r ) / ξ c , where ξ t is the maximum value of the quasi-static response M m of the vehicle flow load under the action of the test vehicle fleet during open traffic and moving loading, and ξ rIt is the average value of the quasi-static response of the vehicle flow load when there is no vehicle on the bridge deck or only a car after the test vehicle fleet drives off the bridge deck, ξ c It is the theoretical calculated load effect R of the corresponding test vehicle fleet and environmental vehicles m Maximum value

[0142] Perform dynamic time warping on the theoretical calculated response and the quasi-static action of the vehicle flow load to align their sequences. The steps are as Figure 10 shown and specifically include

[0143] Step S551: Construct the Euclidean distance matrix ZT×T of the two time series

[0144] Step S552: Recursively construct the cumulative distance matrix VT×T based on the matrix ZT×T

[0145] Step S553: Find the numerically minimum path (the shortest path) from the upper left corner to the lower right corner of the cumulative distance matrix VT×T, which is the optimal warping path Kload of the monitored response sequence and the calculated theoretical response sequence

[0146] Step S554: Delete the sequence duplicates and fine-tune the spatio-temporal trajectories of individual vehicles according to the shortest distance path to achieve fine-tuning matching between the response time series

[0147] The calculation formula for the structural check coefficient ψ of is ψ = (ξ t - ξ r ) / ξ c , where ξ t is the maximum value of the quasi-static response M of the vehicle flow load under the moving load of the test vehicle fleet during traffic opening, ξ m is the average value of the quasi-static response of the vehicle flow load when there is no vehicle on the bridge deck or only a car after the test vehicle fleet drives off the bridge deck, ξ r is the theoretical calculated load effect R of the corresponding test vehicle fleet and environmental vehicles c Maximum value. According to the value range of the structural check coefficient, the operating status of the bridge and its health monitoring system is judged as shown in Table 1 below. The judgment of the bridge operating status mainly depends on the value of the structural check coefficient at the key section measuring points. For the bridge monitoring system, the data accuracy and effectiveness of each measuring point are judged according to the structural check coefficient of each measuring point. When the operating status is judged to be abnormal, it can be judged that the monitoring system is abnormal or the monitoring system is normal while the bridge status is abnormal m Table 1. Different structural check coefficients reflect the operating status of the bridge and its monitoring system

[0148] Table 1. Different structural check coefficients reflect the operating status of the bridge and its monitoring system

[0149]

[0150]

[0151] The matching front and rear effects are as follows Figure 11 shown. It can be seen that compared with Figure 11 (a) Fine-tune the peak values of the measured and calculated deflections of the vehicle flow load in the front vehicle, Figure 11 (b) After fine-tuning and matching, the peak values of the measured and calculated deflections of the vehicle flow load are more accurately matched; compare the theoretical and measured response amplitudes after dynamic time warping, calculate the structural verification coefficient, and simultaneously evaluate the bridge structure state and the accuracy of the monitoring system; The matching effects of each monitoring point of the bridge in the embodiment are as described in the following table:

[0152] Table 2: Matching effects of each monitoring point of the bridge in the embodiment

[0153]

[0154] As can be seen from the table, the inventive method has been implemented and verified on the bridge in the embodiment. The test loading efficiency of the test vehicle fleet meets the threshold value of 0.7; the deflection verification coefficients of each measuring point of the structure are all between 0.50 and 0.90, and it can be judged that the monitoring system and the bridge state are good.

[0155] Embodiment 2:

[0156] As Figure 12 shown, this embodiment provides a traffic detection system that does not interfere with the operating state of a bridge and its monitoring system, which includes four components: an artificial test vehicle fleet subsystem, a vehicle spatio-temporal trajectory acquisition subsystem, a vehicle trajectory coordinate mapping subsystem, and a bridge and its monitoring system evaluation subsystem;

[0157] The artificial test vehicle fleet subsystem is the carrier of the detection, which is composed of a number of trucks of the same model according to the designed loading scheme. The designed loading scheme includes the number of trucks, total weight, axle weight, wheelbase, following distance, lane arrangement, driving speed, etc.; The loading of the designed loading scheme can reach the load test efficiency threshold value determined by the bridge load test specification;

[0158] The vehicle spatio-temporal trajectory acquisition subsystem obtains the driving trajectory of the artificial test vehicle fleet under the vision of the unmanned aerial vehicle and the dynamic world coordinates recorded by the RTK of the leading vehicle; The driving trajectory of the artificial test vehicle fleet under the vision of the unmanned aerial vehicle is obtained by using the machine vision algorithms YOLOv11 and ByteTrack to identify the driving trajectories of all vehicles on the bridge surface during the process of the test vehicle fleet crossing the bridge;

[0159] The vehicle trajectory coordinate mapping subsystem calculates the dynamic world coordinates of all vehicles crossing the bridge by using the dynamic time warping algorithm for the information obtained by the position trajectory acquisition subsystem; The dynamic time warping algorithm infers the cumulative distance matrix by constructing the Euclidean distance matrix between the RTK vehicle trajectory of the leading vehicle and the trajectory of the leading vehicle identified by YOLOv11 and ByteTrack, and finds the optimal warping path of the cumulative distance matrix;

[0160] The vehicle trajectory coordinate mapping subsystem also loads the dynamic world coordinates of all vehicles crossing the bridge into the finite element model of the bridge to obtain the theoretical vehicle flow load response of the monitoring points;

[0161] The bridge and its monitoring system evaluation subsystem calculates the structural check coefficient through the dynamic programming theory and the measured vehicle flow load response sequence. The dynamic programming theory infers the cumulative distance matrix by constructing the Euclidean distance matrix of the monitoring data vehicle flow load action response time sequence and the corresponding calculated theoretical response time sequence, and finds the optimal warping path of the monitoring response sequence and the calculated theoretical response sequence;

[0162] The bridge and its monitoring system evaluation subsystem also determines the operating state of the bridge and its monitoring system according to the calculated value of the structural check coefficient.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the operating state of a bridge and its monitoring system without disturbing traffic, characterized in that, The method comprises the following steps: S1. Determine the load test efficiency threshold value according to the bridge load test specification, and set the test vehicle fleet queue parameters according to this threshold value, including the number of trucks, total weight, axle weight, wheelbase, following distance, lane arrangement, and driving speed; S2. Drive the test vehicle fleet across the bridge according to the fleet loading plan. An RTK positioning system is arranged on a leading truck in the test vehicle fleet to record its dynamic world coordinates during driving on the bridge deck. The unmanned aerial vehicle hovers over the bridge to aerial photograph the whole process video of the test vehicle fleet passing through the test bridge, and the health monitoring system records various monitoring data during the process of the test vehicle fleet passing through the bridge; S3. Use the YOLOv11 and ByteTrack algorithms to identify the driving trajectories of all vehicles on the bridge deck during the process of the test vehicle fleet passing through the bridge. According to the dynamic world coordinates of the leading truck equipped with the RTK system and its trajectory information in the unmanned aerial vehicle camera view, establish a mapping relationship by using the dynamic time warping algorithm; S4. Apply the mapping relationship established in step S3 to the video recognition trajectory information of other vehicles passing through the bridge in the test vehicle fleet, obtain the dynamic world coordinates of all vehicles in the test vehicle fleet during driving on the bridge deck, and load them into the bridge finite element model to obtain the theoretical responses of each monitoring point of the bridge; S5. Separate the signals of the monitoring data recorded by the health monitoring system described in step S2 to obtain the measured response components under the action of the vehicle flow load, perform dynamic time warping with the theoretical responses obtained in step S4 to align their sequences, compare the amplitudes, calculate the structural calibration coefficient, and simultaneously evaluate the bridge structure state and the monitoring system accuracy.

2. The method for detecting the operating state of the bridge and its monitoring system without disturbing traffic according to claim 1, characterized in that The load test efficiency threshold value for step S1 is η min , and the load test efficiency η of the test vehicle fleet is calculated by the formula η = L a / [L d (1 + μ)], where L a is the maximum calculated load effect of the test vehicle fleet crossing the bridge, L d is the maximum calculated load effect of the vehicle load model in the bridge design code, and μ is the vehicle load impact factor calculated in the bridge design code; by adjusting the queue parameters of the test vehicle fleet, the load test efficiency η of the test vehicle fleet is made to satisfy η ≥ η min , and the queue parameters of the test vehicle fleet include the number of trucks, total weight, axle weight, wheelbase, following distance, lane arrangement, and driving speed.

3. The method for detecting the operating state of the bridge and its monitoring system without interfering with traffic according to claim 1, characterized in that The YOLOv11 algorithm in step S3 is used to identify the vehicle contour information through instance segmentation from the perspective of the unmanned aerial vehicle, specifically including: Step 311: Use a drone to pre-take vehicle images from multiple angles as a training dataset. Precisely annotate the vehicle contour area in each image, and record the center point coordinates (x cl , y cl ) and width and height (w l , h l ) of the annotation rectangle. Randomly divide the annotated dataset into a training set, a test set, and a validation set at a ratio of 80%, 10%, and 10%; Step 312: Train the YOLOv11 network model with the training set. In each batch training, based on the YOLO model, predict the coordinates (x cp , y cp ) of the center point of the bounding box in the vehicle area of the image and the width and height (w p , h p ), and calculate the corresponding location loss classification loss and confidence classification loss comprehensive loss The calculation formulas are as follows: where λ coord is the weight coefficient for weighing different losses; S is the number of grids of the feature map extracted by the model; B is the number of bounding boxes predicted for each grid; indicates whether the j-th bounding box in the i-th grid contains the target, being 1 if it does and 0 otherwise; x cp,i , y cp,i are the center coordinates of the bounding box of the vehicle area in the i-th grid of the image predicted by the YOLO model; x cl,i , y cl,i are the center coordinates of the bounding box of the vehicle area in the i-th grid of the image annotation; w p,i , h p,i are the width and height of the bounding box of the vehicle area in the i-th grid of the image predicted by the YOLO model; w l,i , h l,i are the width and height of the bounding box of the vehicle area in the i-th grid of the image annotation; C represents the label of the vehicle contour; indicates whether the i-th grid contains the target, being 1 if it does and 0 otherwise, p i (C) is the confidence of the bounding box of the label category C in the i-th grid predicted by the model; is the true confidence of the target category in the i-th grid; M is the number of samples in the training set; y mc is the true confidence of the c-th category of the m-th sample; α is the positive and negative sample balance coefficient; γ is the small sample focusing coefficient; classes is the label of the vehicle contour area; Calculate the comprehensive loss through backpropagation Gradient of weight W Update the model weights of the corresponding training batch through the Adam optimizer: where ρ is the generalized learning rate, and W q+1 is the model weight of the (q + 1)-th batch; W q is the model weight of the q-th batch; Step 313: Record the comprehensive loss of each training batch and the model weights W, and find the weights that minimize the comprehensive loss through loop training, and denote the optimal weights as W optmal . Use the YOLOv11 model configured with the optimal weights W optmal to identify the images in the whole-process video of the test vehicle convoy passing through the test bridge during the aerial photography of the drone hovering over the bridge obtained in step S2, and identify the center point coordinates (x ca,n , y ca,n ) and width-height (w a,n , h a,n ) of the bounding boxes of the outline, wheels and license plate area of the nth vehicle, and complete the instance segmentation of the vehicle contour information.

4. The method for detecting the operating state of a bridge and its monitoring system without interfering with traffic according to claim 3, wherein The ByteTrack algorithm described in step S3 is used to track the targets of the vehicles to obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle fleet passing through the bridge, specifically including: Step 321: Based on the YOLOv11 model with the optimal configuration weight W optmal , identify the nth vehicle in the t-th frame image of the test vehicle convoy crossing the bridge, and extract the center point coordinates (x ca,n,t , y ca,n,t ) of its contour bounding box as trajectory points; Step 322: Input the center point coordinates (x ca,n,t , y ca,n,t ) of the nth vehicle contour bounding box in the tth frame into the ByteTrack object tracking algorithm, and use Kalman filtering to predict the center point coordinates (x ck,n,t , y ck,n,t ) of the vehicle in the tth frame: where, F t is the state transition matrix; B t is the control input matrix; v t is the control vector; P t|t-1 is the prediction covariance matrix; Q t is the process noise covariance matrix; (x cf,n,t-1 , y cf,n,t-1 ) is the corrected trajectory state updated by Kalman filtering in the (t - 1)-th frame, the superscript T represents the transpose of the matrix, and P t-1 represents the prediction covariance matrix in the (t - 1)-th frame; Step 323, calculate the intersection over union (IoU) of the bounding box (x ca,n,t ±w a,n , y ca,n,t ±h a,n ) and the predicted bounding box (x ck,n,t ±w a,n , y ck,n,t ±h a,n ), which is the ratio of the area of intersection (AoI) to the area of union (AoU) of the bounding boxes: where AoI1 and AoI2 respectively represent the overlapping lengths of the two bounding boxes in the horizontal and vertical directions; h a,n represents the width and height of the contour bounding box of the nth vehicle; Step 324: For the detection boxes with IoU values exceeding the set threshold IoU limit , update the corrected trajectory state (x cf,n,t , y cf,n,t ) of the nth vehicle in the tth frame through Kalman filtering: where K t is the Kalman gain matrix; H t is the observation matrix; R t is the observation noise covariance matrix, and I represents the identity matrix; Step 325: Delete vehicle trajectories that have not been matched for more than 5 seconds, and output the matching tracking results of the nth vehicle, including the vehicle ID, vehicle speed (calculated by the difference in the center point coordinates of the front and rear frames), and vehicle trajectory {(x cf,n,t , y cf,n,t )|t = 0, 1, 2,...}; Loop through steps 321 to 325 to obtain the spatio-temporal trajectory information of all vehicles on the bridge deck during the entire process of the test fleet crossing the bridge.

5. The method for detecting the operating state of a bridge and its monitoring system without disturbing traffic according to claim 4, characterized in that The dynamic time warping algorithm described in step S3 is based on the dynamic world coordinates of the leading truck installed with the RTK system, and establishes a conversion mapping relationship from the trajectory coordinates of the leading truck's UAV captured images to the world coordinates recorded by the RTK. This conversion relationship is the optimal warping path S that enables the dynamic time warping algorithm to match the two time coordinate sequences load , specifically including: Step 331: Record the world coordinates of the leading truck equipped with the RTK system as the RTK vehicle trajectory of the leading truck \(\{(x crtk,l,t , y crtk,l,t )|t = 0, 1, 2,...\}. By resampling, make the time series lengths of the leading truck drone captured images, the leading vehicle trajectory identified by the YOLOv1 1 model and the ByteTrack algorithm \(\{(x cf,l,t , y cf,l,t )|t = 0, 1, 2,...\) consistent, and construct the Euclidean distance two-dimensional matrix D L×L of \(\{(x crtk,l,t , y crtk,l,t )|t = 0, 1, 2,...\) and \(\{(x cf,l,t , y cf,l,t )|t = 0, 1, 2,...\): In the formula, u is the index of the RTK trajectory sequence; g is the index of the unmanned aerial vehicle trajectory sequence; Step 332: Based on the matrix D L×L Recursively construct the cumulative distance matrix C L×L : C L×L (u, g) = D L×L (u, g) + min{C(u - 1, g), C(u, g - 1), C(u - 1, g - 1)} Finding the cumulative distance matrix C L×L The numerical minimum path (shortest path) from the upper left corner to the lower right corner is the optimal warping path S for the matching of the RTK and UAV trajectory sequences load .

6. The method for detecting the operating state of a bridge and its monitoring system without disturbing traffic according to claim 5, characterized in that The application of the mapping relationship established in step S3 to the video recognition trajectory information of other vehicles passing through the bridge in the test vehicle fleet described in step S4 specifically includes: Step 411: Based on the spatio-temporal trajectory information of all vehicles on the bridge deck during the whole process of the test vehicle fleet crossing the bridge, construct a vehicle identification trajectory x-coordinate matrix X video,Z×K and a y-coordinate matrix Y video,Z×K : In the matrix, K is the total number of vehicles in the bridge traffic flow, where k = 1 is the leading vehicle; Z is the time span from the leading vehicle getting on the bridge to the last vehicle of the convoy getting off the bridge; and respectively represent the trajectory x coordinate and the trajectory y coordinate of the k-th vehicle at time z; Step 412: Based on the obtained optimal distortion path S load , synchronously adjust the sequence duplicates in the x and y coordinate matrices of the vehicle recognition trajectory, and adjust the number of matrix rows to regularize the overall coordinate of the bridge deck traffic flow, so as to obtain the x and y world coordinate matrices X real,Z×K , Y real,Z×K :

7. The method for detecting the operating state of a bridge and its monitoring system without disturbing traffic according to claim 6, characterized in that, The obtaining of the dynamic world coordinates of all vehicles in the test vehicle fleet during driving on the bridge deck and the loading into the bridge finite element model to obtain the theoretical responses of each monitoring point of the bridge described in step S4 specifically includes: Step 421: Based on the test vehicle fleet queue parameters formulated in step S1, and combined with the x-world coordinate matrix X of all vehicles crossing the bridge obtained in step 72 real,Z×K and the y-world coordinate matrix Y real,Z×K , construct a multivariate traffic flow load dynamic time series S(z): In the formula, are the x and y world coordinates of the k-th vehicle at time z; b k is the number of axles of the k-th vehicle; is the distance between the x coordinate of the trajectory and the y coordinate of the trajectory of the β-th axle relative to time z; is the axle weight load of the β-th axle; Step 422: Extract the influence surface function I of each monitoring point of the bridge, i.e., the displacement, deflection, and strain measurement points m =(x b , y b ), where m is the monitoring point number, and I m =(x b , y b ) represents the response value of the measurement point m when a unit concentrated load is applied at the world coordinates (x b , y b ) of the bridge deck; based on the multivariate traffic flow load dynamic time series constructed in Step 81, calculate the theoretical response time series R m (z): where r z is the calculated theoretical response value of measuring point m at time z, including displacement, deflection, and strain; is the x - coordinate of the β - axis of the k - th vehicle on the bridge deck at time z; is the y - coordinate of the i - th axis of the k - th vehicle on the bridge deck at time t.

8. The method for detecting the operating state of a bridge and its monitoring system without interfering with traffic according to claim 1, characterized in that, The separation of the signals of the monitoring data recorded by the health monitoring system described in step S2 to obtain the measured response components under the action of the vehicle flow load and the performance of dynamic time warping with the theoretical responses obtained in step S4 to align their sequences described in step S5 specifically includes: Step 51. Resample the monitoring data traffic load response time series M of the measuring point m m ={m z |z = 0, 1, 2,...} to be of the same length as the calculated theoretical response time series R m ={r z |z = 0, 1, 2,...} of the measuring point m, and construct the Euclidean distance matrix O m of M m and R Z×Z : In the formula, is the index of the monitored response sequence; ∈ is the index of the calculated theoretical response sequence; Step 52, based on the matrix O Z×Z Recursively construct the cumulative distance matrix V Z×Z : Finding the cumulative distance matrix V Z×Z The numerical minimum path (shortest path) from the upper left corner to the lower right corner is the optimal warping path A between the monitoring response sequence and the calculated theoretical response sequence load ; Delete sequence duplicates and adjust R m The sequence length of R is adjusted to achieve fine-tuning matching between response time sequences The calculation formula for the structural verification coefficient ψ of step S5 is ψ = (ξ t - ξ r ) / ξ c , where ξ t is the maximum value of the quasi-static response M m of the vehicle flow load under the moving load of the test vehicle fleet during open traffic, ξ r is the average value of the quasi-static response of the vehicle flow load when there is no vehicle or only a car on the bridge deck after the test vehicle fleet has left the bridge deck, and ξ c is the maximum value of the theoretical calculated load effect R m corresponding to the test vehicle fleet and environmental vehicles.

9. A bridge and its monitoring system for detecting the operating state without disturbing the traffic detection system, characterized in that, The system is loaded with the method described in any one of claims 1-8. The system includes four components: an artificial test vehicle fleet subsystem, a vehicle spatio-temporal trajectory acquisition subsystem, a vehicle trajectory coordinate mapping subsystem, and a bridge and its monitoring system evaluation subsystem. The artificial test vehicle fleet subsystem is the carrier for detection, consisting of a number of trucks of the same vehicle type according to the designed loading scheme. The vehicle spatio-temporal trajectory acquisition subsystem obtains the driving trajectory of the artificial test vehicle fleet under the vision of the unmanned aerial vehicle and the dynamic world coordinates recorded by the RTK of the leading vehicle. The vehicle trajectory coordinate mapping subsystem uses the dynamic time warping algorithm to calculate the dynamic world coordinates of all vehicles crossing the bridge from the information obtained by the position trajectory acquisition subsystem, and loads the dynamic world coordinates of all vehicles crossing the bridge into the finite element model of the bridge to obtain the theoretical traffic flow load response of the monitoring points. The bridge and its monitoring system evaluation subsystem calculates the structural verification coefficient through the dynamic regularization theory and the measured traffic flow load response sequence, and determines the operating state of the bridge and its monitoring system according to the calculated value of the structural verification coefficient.

Images