A Method and System for Measuring Lateral Displacement of Bridge Piers After Ship Collision Based on Multi-Source Data Fusion

Abstract

This invention provides a method and system for measuring the lateral displacement of a bridge pier after a ship collision based on multi-source data fusion. The method includes: acquiring sensor observation data and calculating the bridge sensor displacement based on the sensor observation data; acquiring a set of bridge image frames, extracting the keypoint set corresponding to each bridge image frame set using the SuperPoint model, and then completing the keypoint set; calculating the bridge pixel displacement sequence based on the keypoint set, and then converting the bridge pixel displacement sequence to physical space to obtain the bridge visual displacement sequence; fusing the bridge visual displacement sequence and the bridge sensor displacement sequence using the EKF algorithm to obtain the observed bridge displacement; and fusing the bridge observed displacements corresponding to all cameras to obtain the actual bridge displacement. This invention solves the problem in existing technologies where measurement using visual methods is easily affected by environmental factors such as rain, fog, and dust, as well as object occlusion, leading to a decrease in measurement accuracy.

Description

Technical Field

[0001] This invention relates to the field of bridge collision prevention and monitoring technology, and in particular to a method and system for measuring the lateral displacement of bridge piers in a ship collision based on multi-source data fusion. Background Technology

[0002] Bridge structural health monitoring is a crucial means of ensuring traffic safety and the long-term use of bridges. Ship impacts on bridge piers are a typical example of sudden dynamic load events that directly affect the safety, load-bearing capacity, and service life of the pier structure. Therefore, accurate and real-time acquisition of pier displacement information after impact is of great significance for emergency response, structural damage assessment, and safety decision-making. However, current technologies still have several limitations in this area.

[0003] Visual methods are commonly used for measurement. For example, laser ranging and lidar technologies measure the distance to target points by emitting laser beams onto the surface of bridge piers, enabling 3D modeling or deformation monitoring. Displacement measurement methods based on monocular video images capture video of bridge piers using cameras, calculate pixel displacement using digital image correlation techniques or feature point tracking algorithms, and convert it into actual physical displacement. While these methods are widely used in bridge structure monitoring, they have limitations in dynamic monitoring scenarios involving ship collisions. Both laser signals and camera footage rely on received light signals for judgment. However, light signals are easily affected by environmental factors such as rain, fog, and dust, resulting in scattering or attenuation, leading to blurred or missing images and decreased measurement accuracy. Furthermore, laser equipment and camera shake, as well as ship hull obstruction, can easily cause the loss of identified feature points or tracking drift, affecting displacement continuity and accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion. This method solves the problem that existing technologies using visual methods are easily affected by environmental factors such as rain, fog, and dust, as well as object obstruction, resulting in blurred or missing images, which leads to a decrease in measurement accuracy.

[0005] According to an embodiment of the present invention, a method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion includes: Acquire sensor observation data, and then calculate the bridge sensor displacement based on the sensor observation data; Acquire a set of bridge image frames captured by multiple cameras, use the SuperPoint model to extract the set of key points corresponding to each bridge image frame set, and then complete the key point set. Based on the key point set, the bridge pixel displacement sequence is calculated, and then the bridge pixel displacement sequence is converted into physical space to obtain the bridge visual displacement sequence. The bridge's visual displacement sequence and bridge sensor displacement sequence are fused using the EKF algorithm to obtain the observed bridge displacement. The bridge displacements observed by all cameras are fused to obtain the actual bridge displacement.

[0006] Preferably, the key point set includes multiple key points and a search area, and each key point contains a point state and a descriptor vector; Methods for completing keypoint sets include: The keypoint set is cleaned, and then for each keypoint in the keypoint set, a corresponding state transition equation is established. Based on the point state of each keypoint in the current frame, the state transition equation is used to predict the predicted position of the keypoint in the next frame. In response to the existence of the predicted position of the key point in the next frame in the key point set, the visual observation vector of the key point in the next frame is calculated based on the point state of the key point in the current frame. If the predicted position of a key point in the next frame does not exist in the key point set of the next frame, the sensor observation displacement increment for the next frame is extracted from the sensor observation data, and then the sensor observation displacement increment is projected onto the pixels to obtain the pixel displacement. Calculate the sensor observation vector of key points in the next frame based on pixel displacement; Based on the visual observation vector or sensor observation vector, the extended Kalman prediction algorithm is used to calculate the position of the key point at the next moment and update the key point set for the next frame.

[0007] Preferably, the method for cleaning the set of key points includes: Based on the point states and corresponding descriptor vectors of key points in any two adjacent frames, calculate the matching confidence between all key points in the two frames, and remove key points with matching confidence greater than a preset value. Based on the point states of keypoints in any two adjacent frames, the RANSAC algorithm is used to verify the geometric consistency of each keypoint and remove all keypoints that do not meet the geometric consistency requirement.

[0008] Preferably, the calculation formula for the extended Kalman prediction algorithm is as follows: in, This indicates the key point location for the next frame. The predicted location of the key point in the next frame. Let P be the visual observation vector or the sensor observation vector, where P is the matching probability and R is the observation noise.

[0009] Preferably, for any keypoint in the keypoint set, if the predicted position of the keypoint in the next two consecutive frames is not in the keypoint set, the state transition equation is used to predict the predicted position of the keypoint in the next frame based on the position of the keypoint in the current frame. Then, based on the predicted position and the search area of ​​the keypoint set, the SuperPoint model is used to construct the search area of ​​the next frame. Finally, the SuperPoint model is used to re-detect within the search area of ​​the next frame.

[0010] Preferably, a key point sequence is determined based on the key point set, then the cumulative displacement between any two adjacent key points in the key point sequence is calculated, and the cumulative displacement is combined into a bridge pixel displacement sequence.

[0011] Preferably, the method for fusing the bridge observation displacements corresponding to all cameras to obtain the actual bridge displacement includes: Transform all bridge observation displacements to the global coordinate system to obtain the global observation displacements; The actual displacement of the bridge is obtained by adaptively weighting and fusing all global observed displacements.

[0012] Preferably, the calculation formula for adaptive weighted fusion is as follows: in, The weight (0~1) is the weight of the k-th global observation displacement. This is the weight sensitivity coefficient (generally taken as 0.1~5). The comprehensive error of the k-th globally observed displacement is... is the comprehensive error index for the j-th globally observed displacement.

[0013] On the other hand, according to embodiments of the present invention, a system for measuring the lateral displacement of a ship-collision bridge pier based on multi-source data fusion is also provided. This system uses the aforementioned method for measuring the lateral displacement of a ship-collision bridge pier based on multi-source data fusion, and includes: The data acquisition module includes a camera array and a sensor array. The camera array is used to acquire a set of bridge image frames, and the sensor array is used to acquire sensor observation data. A visual calculation module is used to calculate the bridge pixel displacement sequence and then convert the bridge pixel displacement sequence into physical space to obtain the bridge visual displacement sequence. The matching module is used to calculate the bridge pixel displacement sequence using the SuperGlue matching algorithm, and then convert the bridge pixel displacement sequence into physical space to obtain the bridge visual displacement sequence. A sensor calculation module is used to calculate the bridge sensor displacement based on sensor observation data. The displacement calculation module is used to fuse the bridge visual displacement sequence and the bridge sensor displacement sequence using the EKF algorithm to obtain the bridge observation displacement, and then fuse the bridge observation displacements corresponding to all cameras to obtain the actual bridge displacement.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention uses multiple cameras to acquire a set of bridge image frames and employs the SuperPoint deep network for feature extraction. It extracts key points and corresponding descriptor vectors from each image in the bridge image frame set, then completes the key points to overcome the influence of object occlusion. The SuperGlue matching algorithm is used to track key points in consecutive image frames, avoiding image blurring caused by environmental factors such as rain, fog, and dust. Simultaneously, combined with sensor observation data, the EKF algorithm is used to calculate the bridge's observed displacement, significantly improving the accuracy and continuity of displacement measurement. Attached Figure Description

[0015] Figure 1 This is a diagram illustrating the lateral displacement measurement method according to an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] like Figure 1 As shown in the figure, this invention proposes a method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion, including: Acquire sensor observation data, and then calculate the bridge sensor displacement based on the sensor observation data; While sensors are slightly inadequate for measuring bridge displacement in sudden events, they offer high accuracy in static measurements. Therefore, this invention employs a sensor array on the bridge to measure the actual displacement of the piers after a ship impact. This invention integrates multiple sensors, including: an IMU that collects the linear acceleration of the pier in the monitoring direction, removes the gravity component, and integrates to obtain the velocity or displacement increment in that direction; a radar sensor that directly outputs the horizontal displacement of the pier; and GPS / GNSS that provides the pier's two-dimensional position and displacement information in a global or engineering coordinate system. All sensor data collected are uniformly converted to the pier coordinate system and projected and normalized along the transverse monitoring direction of interest to form the bridge sensor displacement in that direction. Meanwhile, the sampling data from each sensor is synchronized with the video acquisition using a unified clock or timestamp interpolation method, thus corresponding one-to-one with the bridge image frames acquired by multiple cameras on the time axis. This provides redundant observation for backend fusion and improves robustness in environments with occlusion, changes in lighting, and harsh water conditions.

[0018] Acquire a set of bridge image frames captured by multiple cameras, use the SuperPoint model to extract the key point set and corresponding descriptor vector for each bridge image frame set, and then perform key point completion on the key point set; To avoid problems such as image blurring and missing images caused by shooting angle and environmental influences of a single camera, this invention uses a camera array to capture a set of bridge image frames when a ship collides with a bridge from multiple locations. Each set of bridge image frames contains multiple consecutive bridge image frames.

[0019] For each bridge image frame in the bridge image frame set, each frame image Inputting the SuperPoint feature extraction network, stable keypoints are selected through non-maximum suppression, resulting in the set of keypoints in the t-th frame image. And generate a corresponding descriptor vector for each key point. , is a dimensionless vector. In addition to the descriptor vector, each keypoint also includes the point state (point position and point velocity). At the same time, the SuperPoint model will save the search area corresponding to each keypoint when filtering keypoints. The design of SuperPoint enables it to maintain a high repetition rate and high stability of keypoints in environments such as repetitive textures, low light, and reflective interference.

[0020] For each bridge image frame's keypoint set, due to various types of image noise, the keypoint set will inevitably contain some erroneous keypoints, therefore it is necessary to... Use SuperGlue to create descriptor vectors for key points. and point position As input, contextual information is propagated through a graph neural network (GNN) to calculate the matching probability of keypoint pairs formed by keypoints between adjacent frames; in It is the learned matching mapping function. It is the descriptor vector of the j-th keypoint in the set of keypoints corresponding to frame t+1.

[0021] Matching confidence To measure the reliability of each keypoint pair, SuperGlue, with its built-in global geometric consistency and attention mechanism, can directly output high-confidence matches in most scenarios, and then remove all keypoints with a match confidence greater than a preset value.

[0022] To further improve the geometric consistency of the matching results, this invention introduces RANSAC to remove abnormal matches based on the SuperGlue matching results, and finally obtains a set of key points that are all normal key points.

[0023] To track the motion of key points over time, adjacent frames are... and The keypoint set is matched using SuperGlue. Due to complex factors such as occlusion, water reflection, low light, and motion blur in the bridge area, as well as the removal of some abnormal keypoints, the SuperPoint model may fail to detect some keypoints in several frames, resulting in trajectory interruption and discontinuity. Therefore, it is necessary to complete the keypoints in the keypoint set first.

[0024] For each keypoint i in a set of keypoints, construct a four-dimensional state vector at time t (frame t): in, and These are the horizontal and vertical coordinates (point positions) of the key points in the pixel plane. and These are the corresponding horizontal and vertical pixel velocities (dot velocities).

[0025] At high frame rates, keypoints can be approximated as moving at a uniform velocity, and their state transition equation is: in, This refers to the inter-frame time interval. The process noise is used to characterize the deviation between the actual motion of key points and the ideal uniform velocity model, such as that caused by acceleration perturbations, local non-uniform motion, etc.; Q is the process noise covariance matrix, used to describe uncertainties such as acceleration perturbations and local non-uniform motion, and its recommended form is: in, The acceleration noise variance reflects the magnitude of acceleration fluctuations at key points.

[0026] Based on the state transition equation, the predicted location of the keypoint in frame t+1 can be calculated. If a keypoint is successfully detected in the keypoint set corresponding to frame t+1, then the visual observation is: Its update equation is: in, Let be the visual observation vector of the i-th keypoint in frame t+1; Visual observation noise represents the measurement uncertainty caused by factors such as image resolution, feature detection, and matching errors. The covariance matrix of visual observation noise, the size of which can be determined by the SuperGlue matching probability. Make dynamic adjustments: in, The baseline covariance matrix for visual observation noise (in pixels). 2 ), representing the matching probability The observed noise level at that time; This represents the variance of the pixel error in visual observation.

[0027] If no keypoint is detected in the keypoint set corresponding to frame t+1, then the sensor observation displacement increment for the next frame is extracted using the sensor observation data. , And it must undergo camera extrinsic parameter transformation. Transform to the camera coordinate system: Utilizing camera internal parameters Projection is pixel displacement: The final sensor observation vector of the i-th keypoint in frame t+1 is: Whether using visual observation or sensor observation, the same extended Kalman prediction algorithm is used to calculate the position of keypoint i in frame t+1. : in, Let be the covariance matrix of the sensor observation noise. Let Variance be the variance of the sensor's lateral displacement observation error. Let be the variance of the sensor's vertical displacement observation error, and P be the matching probability.

[0028] Then, the position of this key point is added to / updated in the key point set corresponding to frame t+1.

[0029] If keypoint i is not visually detected for two consecutive frames, relying solely on extended Kalman prediction will lead to a gradual accumulation of prediction errors. To improve the stability of trajectory recapture, a region of interest (ROI) for the next frame is constructed at each time t using the predicted positions of the keypoints. Let the keypoint at time t+1 be obtained from the state prediction. Corresponding point position Then the ROI used for SuperPoint re-detection in frame t+1 is defined as: in, The radius of the ROI at time t is determined by the search region corresponding to the set of key points at time t, and its adaptive update is as follows: here, The minimum search radius (in pixels). The predicted velocity vector for key points. For its modulus length, This is the variance term (in pixels) in the state covariance matrix corresponding to the horizontal pixel position component. , The weighting coefficients are used. This adaptive ROI strategy expands the search area as the keypoint's movement speed increases, and automatically widens the search range when the state covariance expands. This allows SuperPoint to quickly re-detect and match within the ROI when the keypoint re-enters the field of view.

[0030] Based on the key point set and the corresponding descriptor vector, the bridge pixel displacement sequence is calculated, and then the bridge pixel displacement sequence is converted into physical space to obtain the bridge visual displacement sequence. After completing the keypoint set for each frame, calculate the cumulative displacement of keypoints between adjacent frames: Horizontal direction: Vertical direction: Then, all the cumulative displacements are combined into a bridge pixel displacement sequence.

[0031] Based on the known geometric dimensions of the bridge piers and the camera's internal parameters, it is necessary to accumulate the displacement of key points. This invention converts physical displacements to engineering scales using a perspective projection model. The cumulative physical displacement of the key point in physical space is: in, , Let i be the cumulative displacement of keypoint i in the horizontal and vertical directions in the image coordinate system. This represents the accumulated amount of the corresponding physical displacement. , This represents the equivalent focal length in the horizontal and vertical directions within the camera's intrinsic parameter matrix. The depth distance of key point i along the optical axis in the camera coordinate system can be directly measured using equipment such as laser rangefinders.

[0032] Using Euclidean algorithm, the visual displacement of the bridge between two adjacent frames can be obtained. Then, all the bridge visual displacements can be combined into a bridge visual displacement sequence. .

[0033] Through the above transformation, pixel-domain motion quantities can be uniformly mapped to physical displacements at the engineering scale, providing input data for the impact response analysis of bridge piers. The bridge's visual displacement sequence and bridge sensor displacement sequence are fused using the EKF algorithm to obtain the observed bridge displacement. bridge visual displacement sequence With sensor displacement sequence The EKF input is fused to achieve time-series filtering and obtain the observed bridge displacement. .

[0034] EKF uses a state prediction + observation update mechanism to suppress noise, resulting in a continuous, smooth, and stable final displacement sequence. EKF automatically adjusts the influence of vision and sensors based on the noise matrix weights. EKF residuals are also used to guide front-end denoising and ROI adjustment, forming a cross-module closed loop.

[0035] The bridge displacements observed by all cameras are fused to obtain the actual bridge displacement.

[0036] Since there are multiple cameras, multiple sets of bridge observation displacements will be obtained. All bridge observation displacements will be transformed to the global coordinate system: in, Let be the rotation matrix from the k-th camera coordinate system to the global coordinate system; This represents the k-th bridge observation displacement sequence in pixel coordinates. This represents the k-th bridge observation displacement sequence in the global coordinate system.

[0037] Then, based on the adaptive weighted fusion of observation quality, the final actual displacement of the bridge is obtained. : in, The weight (0~1) is the weight of the k-th global observation displacement. This is the weight sensitivity coefficient (generally taken as 0.1~5). The comprehensive error of the k-th globally observed displacement is... This is the comprehensive error index for the j-th globally observed displacement. It is automatically adjusted based on ambiguity, occlusion rate, and matching confidence to make the final displacement more stable.

[0038] This invention uses multiple cameras to acquire a set of bridge image frames and employs the SuperPoint deep network for feature extraction. It extracts key points and corresponding descriptor vectors from each image in the bridge image frame set. After completing the key points, it overcomes the influence of object occlusion and uses the SuperGlue matching algorithm to track key points in consecutive image frames. This avoids image blurring caused by environmental factors such as rain, fog, and dust. Simultaneously, it combines sensor observation data and uses the EKF algorithm to calculate the bridge's observed displacement, significantly improving the accuracy and continuity of displacement measurement.

[0039] On the other hand, embodiments of the present invention also provide a system for measuring the lateral displacement of a ship-collision bridge pier based on multi-source data fusion. This system uses the aforementioned method for measuring the lateral displacement of a ship-collision bridge pier based on multi-source data fusion, including: The data acquisition module includes a camera array and a sensor array. The camera array is used to acquire a set of bridge image frames, and the sensor array is used to acquire sensor observation data. The feature extraction module is used to extract the key point set corresponding to each bridge image frame set using the SuperPoint model, and to complete the key point set. A visual calculation module is used to calculate the bridge pixel displacement sequence and then convert the bridge pixel displacement sequence into physical space to obtain the bridge visual displacement sequence. A sensor calculation module is used to calculate the bridge sensor displacement based on sensor observation data. The displacement calculation module is used to fuse the bridge visual displacement sequence and the bridge sensor displacement sequence using the EKF algorithm to obtain the bridge observation displacement, and then fuse the bridge observation displacements corresponding to all cameras to obtain the actual bridge displacement.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for measuring the lateral displacement of a bridge pier after a ship collision based on multi-source data fusion, characterized in that: include: Acquire sensor observation data, and then calculate the bridge sensor displacement based on the sensor observation data; Acquire a set of bridge image frames captured by multiple cameras, use the SuperPoint model to extract the set of key points corresponding to each bridge image frame set, and then complete the key point set. Based on the key point set, the bridge pixel displacement sequence is calculated, and then the bridge pixel displacement sequence is converted into physical space to obtain the bridge visual displacement sequence. The bridge's visual displacement sequence and bridge sensor displacement sequence are fused using the EKF algorithm to obtain the observed bridge displacement. The bridge displacements observed by all cameras are fused to obtain the actual bridge displacement.

2. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 1, characterized in that: The set of key points includes multiple key points and a search area, and each key point contains a point state and a descriptor vector. Methods for completing keypoint sets include: The keypoint set is cleaned, and then for each keypoint in the keypoint set, a corresponding state transition equation is established. Based on the point state of each keypoint in the current frame, the state transition equation is used to predict the predicted position of the keypoint in the next frame. In response to the existence of the predicted position of the key point in the next frame in the key point set, the visual observation vector of the key point in the next frame is calculated based on the point state of the key point in the current frame. If the predicted position of a key point in the next frame does not exist in the key point set of the next frame, the sensor observation displacement increment for the next frame is extracted from the sensor observation data, and then the sensor observation displacement increment is projected onto the pixels to obtain the pixel displacement. Calculate the sensor observation vector of key points in the next frame based on pixel displacement; Based on the visual observation vector or sensor observation vector, the extended Kalman prediction algorithm is used to calculate the position of the key point at the next moment and update the key point set for the next frame.

3. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 2, characterized in that: Methods for cleaning key point sets include: Based on the point states and corresponding descriptor vectors of key points in any two adjacent frames, calculate the matching confidence between all key points in the two frames, and remove key points with matching confidence greater than a preset value. Based on the point states of keypoints in any two adjacent frames, the RANSAC algorithm is used to verify the geometric consistency of each keypoint and remove all keypoints that do not meet the geometric consistency requirement.

4. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 2, characterized in that: The calculation formula for the Extended Kalman Prediction Algorithm is as follows: in, This indicates the key point location for the next frame. The predicted location of the key point in the next frame. Let P be the visual observation vector or the sensor observation vector, where P is the matching probability and R is the observation noise.

5. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 2, characterized in that: For any keypoint in the keypoint set, if the predicted position of the keypoint is not in the keypoint set in the next two consecutive frames, the state transition equation is used to predict the predicted position of the keypoint in the next frame based on the position of the keypoint in the current frame. Then, based on the predicted position and the search area of ​​the keypoint set, the SuperPoint model is used to construct the search area of ​​the next frame. Finally, the SuperPoint model is used to re-detect within the search area of ​​the next frame.

6. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 2, characterized in that: The key point sequence is determined based on the key point set. Then, the cumulative displacement between any two adjacent key points in the key point sequence is calculated, and the cumulative displacement is combined into a bridge pixel displacement sequence.

7. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 1, characterized in that: Methods for fusing the bridge displacement observations from all cameras to obtain the actual bridge displacement include: Transform all bridge observation displacements to the global coordinate system to obtain the global observation displacements; The actual displacement of the bridge is obtained by adaptively weighting and fusing all global observed displacements.

8. The method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in claim 7, characterized in that: The calculation formula for adaptive weighted fusion is as follows: in, The weight (0~1) is the weight of the k-th global observation displacement. This is the weight sensitivity coefficient (generally taken as 0.1~5). The comprehensive error of the k-th globally observed displacement is... is the comprehensive error index for the j-th globally observed displacement.

9. A system for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion, characterized in that: The system uses a method for measuring the lateral displacement of a ship colliding with a bridge pier based on multi-source data fusion as described in any one of claims 1-8, comprising: The data acquisition module includes a camera array and a sensor array. The camera array is used to acquire a set of bridge image frames, and the sensor array is used to acquire sensor observation data. The feature extraction module is used to extract the key point set corresponding to each bridge image frame set using the SuperPoint model, and to complete the key point set. A visual calculation module is used to calculate the bridge pixel displacement sequence and then convert the bridge pixel displacement sequence into physical space to obtain the bridge visual displacement sequence. A sensor calculation module is used to calculate the bridge sensor displacement based on sensor observation data. The displacement calculation module is used to fuse the bridge visual displacement sequence and the bridge sensor displacement sequence using the EKF algorithm to obtain the bridge observation displacement, and then fuse the bridge observation displacements corresponding to all cameras to obtain the actual bridge displacement.

Images