A method and system for monitoring the settlement displacement of a high-pier large-span continuous rigid frame bridge pier

Abstract

The application belongs to the technical field of bridge intelligent construction, and specifically discloses a high-pier large-span continuous rigid-frame bridge pier settlement displacement monitoring method and system, which comprises the following steps: collecting image sequences at a main target of a pier and at a reference point sub-target under a main girder construction condition; processing the images, including low-illumination image enhancement and high-precision segmentation of optical marker points under a complex background; identifying the optical marker point center coordinates at the main target and the sub-target based on an adaptive threshold gradient Hough transform (ATGHT), extracting the displacement values at the main target of the pier and the reference point sub-target, and then compensating for the camera self-motion measurement error, so as to obtain the pier settlement value of the main girder in the cantilever construction process. The application utilizes a non-contact device to monitor the pier settlement displacement in the main girder construction stage, realizes non-contact, multi-measurement-point, high-precision and high-efficiency monitoring of the pier settlement displacement of the super-high-pier large-span continuous rigid-frame bridge under a complex background, and has the advantages of low cost, convenient use and flexible measurement point selection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pier settlement monitoring, in particular to a pier settlement displacement monitoring method and system for high-pier large-span continuous rigid-frame bridges. BACKGROUND

[0002] With the rapid development of world traffic, bridge construction is gradually moving towards the goal of crossing the sea, crossing the mountainous area, and crossing the international. The pier of large-span bridge, especially the high pier, is an important part of engineering construction, which is related to the quality of the whole bridge construction. One of the key indicators to evaluate the construction quality and safety of the pier is the pier settlement. With the advancement of pier construction and subsequent superstructure construction, the pier will produce settlement deformation. If the settlement value exceeds the allowable range, it will seriously affect the structure linearity and the safety of the bridge. Therefore, the pier settlement needs to be observed during the bridge construction process.

[0003] Through settlement monitoring, abnormal deformation of the bridge can be found in time, and corresponding measures can be taken to prevent harmful damage to the bridge and ensure the safety of the bridge construction. The existing pier settlement monitoring technology can be divided into contact monitoring and non-contact monitoring. The contact monitoring technology refers to directly installing various contact sensors on the bridge body, such as the tensioned wire displacement meter, the vibrating wire embedded strain gauge, and the pressure difference sensor, so as to realize long-time recording of the pier settlement data. However, the existing contact deformation measurement technology has many deficiencies. The high cost of hardware and high sensitivity to noise limit the practical application of contact sensors. In addition, if the pier is located at a high position that is difficult to contact or the pier is located in water, the installation of sensors and the transmission of data are difficult and time-consuming. In order to overcome the various problems faced by contact monitoring, domestic and foreign scholars set up pier settlement observation points and use non-contact devices such as level, total station, GPS positioning system, and microwave radar to monitor the displacement of the settlement observation points to obtain the pier settlement value. In recent years, with the maturity of optical cameras and unmanned aerial vehicles, they have rapidly developed in the field of civil engineering. The combination of computer vision technology and remote cameras provides a promising non-contact solution for bridge structure displacement monitoring and evaluation. However, due to the influence of the complex environment on the measurement, the measurement accuracy of the pier settlement based on computer vision technology is poor. Therefore, the present application provides a non-contact monitoring method and system for high-pier large-span continuous rigid-frame bridge pier settlement under complex environment, which overcomes the deficiencies of the existing pier settlement measurement technology, makes the bridge multi-point deformation measurement based on computer vision technology have high robustness, and solves the problems mentioned in the background technology. SUMMARY

[0004] In order to solve the problems in the prior art, the present application provides a pier settlement displacement monitoring method and system for high-pier large-span continuous rigid-frame bridges, which includes image acquisition, image processing, and error compensation modules. The method can overcome the deficiencies of the existing pier settlement measurement technology, make the bridge multi-point deformation measurement based on computer vision technology have high robustness, and solve the problems mentioned in the background technology.

[0005] To achieve the above object, the present application provides the following technical scheme: a high-pier large-span continuous rigid frame bridge pier settlement displacement monitoring method, comprising the following steps:

[0006] S1, collecting image sequences at the main target and reference point sub-target of the pier under the main girder construction condition;

[0007] S2, processing the images, including low-illumination image enhancement and high-precision segmentation of optical marker points in complex background;

[0008] S3, identifying the optical marker point center coordinates at the main target and sub-target based on the adaptive threshold gradient Hough transform (ATGHT), further extracting the displacement values at the main target and reference point sub-target of the pier, and then compensating the camera self-motion measurement error, i.e. the real displacement of the pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, so as to obtain the pier settlement value of the main girder in the cantilever construction process.

[0009] Preferably, in step S1, specifically comprising:

[0010] S11, placing the dual-camera system on the same tripod and connecting it to the visual controller, the dual-camera system comprising a main camera and a sub-camera;

[0011] S12, pasting optical markers, i.e. main targets, at the pier settlement to be measured, and using the main camera of the dual-camera system to monitor the movement of the main targets deployed on the pier;

[0012] S13, selecting a stationary reference point according to the site conditions, pasting optical markers, i.e. sub-targets, at the reference point, and recording the camera self-motion using the sub-camera of the dual-camera system;

[0013] S14, simultaneously collecting image sequences of the main targets and sub-targets during the construction process of the main girder, including reference images and deformation images.

[0014] Preferably, the low-illumination image enhancement in step S2 specifically improves the image imaging effect to obtain a clear image using the SCI light self-calibration model; the SCI light self-calibration model introduces a weight-shared light learning and self-calibration module, specifically comprising the following:

[0015] Let the light optimization process basic unit be represented as:

[0016]

[0017] Where u t and x t represent the residual term and light at the t stage, respectively, t=0,1,2…,T-1, H θRegardless of the stage number, the light learning with weight sharing is adopted, and the structure H and the parameters θ of the light estimation network are shared in each stage; then, a self-calibration module is introduced, so that the outputs of different stages in the training process can converge to the same state, and the self-calibration module is represented as:

[0018]

[0019] wherein v t is the conversion input of each stage, is a parameterized operator with learning parameters introduced, and a basic unit of the light optimization process is converted as:

[0020] Preferably, the high-precision segmentation of the optical marker points in the complex background in the step S2 is specifically performed by adopting a training network model U 2 -Net based on deep learning. 2 The U 2 -Net is a two-layer nested U-shaped structure and is composed of three parts: a 6-level encoder, a 5-level decoder, and an output fusion structure connected between the decoder and the last level encoder.

[0021] The encoder and the decoder structure include five residual network structures, which are RSU-7, RSU-6, RSU-5, RSU-4 and RSU-4F, wherein RSU-4F uses dilated convolution to replace the up-sampling and down-sampling in the remaining structures; the output fusion structure fuses the saliency maps output by each layer to form the final prediction probability map.

[0022] Preferably, the loss function Train loss of the training network model U 2 -Net is the loss of the saliency maps output by each layer plus the loss of the final fused output prediction probability map, and the calculation formula is as follows:

[0023]

[0024] wherein, is the loss value of the output saliency map, l fuse is the loss value of the final output prediction probability map, and ω fuse are weight coefficients of each part loss, and the standard binary cross-entropy is used to calculate each loss value, and the calculation formula is as follows:

[0025]

[0026] Where (r, c) is the pixel coordinate, (H, W) is the image size, H represents the height, and W represents the width; P G(r,c) and P S(r,c) respectively represent the pixel values of the Ground Truth and the predicted probability map;

[0027] The model U 2 The training evaluation index F β of the Net is calculated according to the following formula:

[0028]

[0029] Where β is a number between 0 and 1, and the training code β 2 is set to 0.3, and the F β value is larger, the better the training effect is;

[0030]

[0031] Where P(r, c) is the predicted probability map, G(r, c) is the corresponding Ground Truth, and MAE is a number between 0 and 1, and the smaller the value is, the better the training effect is.

[0032] Preferably, in step S3, specifically comprising:

[0033] First, the Canny edge detection algorithm is used to perform edge detection on the background removed image output by the U 2 -net network, identify the circular edge, and output a binary image; in the binary image, the pixel gray value at the edge is 1, and the background gray value is 0; the number of pixels with a gray value of 1 in the binary image N is counted, and the pixel radius value of the circle in the reference image and the deformed image is calculated, and the calculation formula is as follows:

[0034]

[0035] Where r is the pixel radius value, n is the number of circular mark points to be detected, and N is the number of pixel points with a gray value of 1 in the binary image;

[0036] Then, a discrete feature curve f(r) is defined on the image gradient field to determine the radius of the circle in the image:

[0037]

[0038] Where q(i, j) is the radial vector from the center of the circle to the pixel (i, j); f(r) is the average value of the dot product of the gradient vector and the radial vector; the pixel (i, j) satisfies the following formula:

[0039]

[0040] wherein, Ar is the interval between adjacent r values;

[0041] According to the characteristic definition that the non-zero gradient vector in the gradient field points to the center of the circle in the gradient Hough circle transform ATGHT, the gradient field is converted into an accumulation array, the gradient intensity value of each pixel of the image edge determines the probability of the center of the circle, the maximum intensity value in the image represents the center of the circle of the optical marker point, and the center coordinates of the optical marker point at the main target and the sub-target are obtained;

[0042] Finally, the center coordinates of the obtained deformed image are subtracted from the center coordinates in the reference image, and the displacement values of the main target and the sub-target are obtained; then, the measurement error of the camera self-motion is compensated, that is, the real displacement of the pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, and the pier settlement value in the main girder construction process is obtained.

[0043] In addition, in order to achieve the above object, the present application also provides the following technical scheme: a high-pier large-span continuous rigid frame bridge pier settlement displacement monitoring system, the system comprises:

[0044] The image acquisition module is used for acquiring the image sequence of the pier main target and the reference point sub-target under the main girder construction condition;

[0045] The image processing module is used for processing the image, including low-illumination image enhancement and high-precision segmentation of the optical marker point under a complex background;

[0046] The error compensation module is used for identifying the center coordinates of the optical marker point at the main target and the sub-target based on the adaptive threshold gradient Hough transform ATGHT, further extracting the displacement values of the pier main target and the reference point sub-target, and then compensating the measurement error of the camera self-motion, that is, the real displacement of the pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, so that the pier settlement value of the main girder in the cantilever construction process is obtained.

[0047] The present application has the advantages that: the present application uses the low-illumination image enhancement based on deep learning, the complex background segmentation algorithm improves the image imaging quality and can accurately segment the optical marker point; the adaptive threshold-based gradient Hough transform is adopted to extract the dynamic displacement of the structure at multiple measuring points and combine the double-camera system to compensate the measurement error caused by the camera self-motion, so that the measurement result is more accurate; the system of the present application comprises the image acquisition, image processing and error compensation modules, forms a complete monitoring system, and the non-contact equipment is used to monitor the pier settlement displacement in the main girder construction stage, so that the non-contact, multi-measuring point, high-precision and high-efficiency monitoring of the pier settlement displacement of the super-high pier large-span continuous rigid frame bridge under a complex background is realized, and the present application has the advantages of low cost, convenient use and flexible selection of measuring points. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1Flow chart of steps of monitoring method in the embodiment;

[0049] Figure 2 Schematic diagram of computer vision-based pier settlement measurement during construction in the embodiment;

[0050] Figure 3 Schematic diagram of computer vision-based pier settlement hardware in the embodiment;

[0051] Figure 4 Schematic diagram of image acquisition in the embodiment;

[0052] Figure 5 Schematic diagram of image processing in the embodiment;

[0053] Figure 6 Schematic diagram of error compensation result in the embodiment;

[0054] Figure 7 Schematic diagram of monitoring system module composition in the embodiment;

[0055] In the figure, 110 is an image acquisition module; 120 is an image processing module; and 130 is an error compensation module. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described in combination with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0057] The present application provides a technical solution: a non-contact monitoring method for pier settlement of a high-pier long-span continuous rigid-frame bridge under complex environment, as shown in Figure 1 The method comprises the following steps:

[0058] I. Image acquisition under main girder construction conditions

[0059] Specifically, the following steps are included:

[0060] 1) Place two sets of industrial cameras on the same tripod and connect them to a visual controller; the double-camera system includes a main camera and a sub-camera;

[0061] 2) Paste optical marker points, i.e. main targets, at the pier settlement to be measured, and use the main camera of the double-camera system to monitor the movement of the main targets arranged on the pier;

[0062] 3) Select a stationary reference point according to the site conditions, paste optical marker points, i.e. sub-targets, at the reference point, and use the sub-camera of the double-camera system to record the camera self-motion.

[0063] 4) Collect the image sequence of the main target and the sub-target of the main beam during the construction process (including the reference image and the deformation image), such as pouring the main beam segment concrete, hoisting the main beam formwork, etc.

[0064] Second, process the image, including low-illumination image enhancement and high-precision segmentation of optical markers in complex background;

[0065] The specific method is:

[0066] 1) The low-illumination image is enhanced by using a deep learning-based method. The SCI illumination self-calibration model is used to improve the image imaging effect. Specifically, the SCI self-calibration illumination learning is a new low-illumination image enhancement model, which introduces a weight-shared illumination learning and self-calibration module.

[0067] The main optimization goal of the model is to accurately calculate the illumination, and then obtain a clear image according to the Retinex theory. The basic unit of the illumination optimization process is:

[0068]

[0069] Where, u t and x t represent the residual term and the illumination of the t(th) stage (t = 0, 1, 2, …, T-1), H θ is irrelevant to the stage number, and the model uses a weight-shared illumination learning, and the illumination estimation network at each stage maintains the structure H and the shared parameters θ. Then the self-calibration module is introduced, so that the outputs of different stages in the training process can converge to the same state, and the self-calibration module can be represented as:

[0070]

[0071] Where, v t (t≥1) is the conversion input of each stage, is a parameterized operator with learning parameters . The basic unit of the illumination optimization process is converted to:

[0072] 2) The image complex background is segmented by using a deep learning-based training network. U 2 -Net is used to remove the invalid background in the image and only keep the region of interest. U 2-Net is a two-layer nested U-shaped structure, which consists of three parts: a 6-level encoder, a 5-level decoder and an output fusion structure connected between the decoder and the last level encoder. The encoder and decoder structure includes 5 residual network structures, respectively RSU-7, RSU-6, RSU-5, RSU-4 and RSU-4F, in which RSU-4F uses dilated convolution to replace the up-sampling and down-sampling in the remaining structures; the fusion structure fuses the output saliency maps of each layer to form the final prediction probability map.

[0073] The loss function Train loss of model training is the loss of the saliency map output by each layer of the network plus the loss of the final fused output prediction probability map, and the specific calculation formula is:

[0074]

[0075] wherein, is the loss value of the output saliency map, l fuse is the loss value of the final output prediction probability map, and ω fuse are the weight coefficients of each part loss. For each loss value, the standard binary cross-entropy is used to calculate, which is:

[0076]

[0077] wherein, (r, c) is the pixel coordinate, (H, W) is the image size, H represents the height, and W represents the width; P G(r,c) and P S(r,c) represent the pixel values of the Ground Truth and the prediction probability map, respectively.

[0078] The calculation formulas of the training evaluation indicators F β and MAE of the model are:

[0079]

[0080] β is a number between 0 and 1, and β 2 is set to 0.3 in the training code. F β (0-1) combines the accuracy and recall, and since the accuracy and recall will get different values according to different probability thresholds, the maxF β is selected as the evaluation indicator, and the larger the value is, the better the training effect is.

[0081]

[0082] wherein, P(r, c) is the prediction probability map, G(r, c) is the corresponding GT (Ground Truth), and MAE is a number between 0 and 1, and the smaller the value is, the better the training effect is.

[0083] III. Camera self-motion measurement error compensation based on adaptive threshold gradient Hough transform and dual-camera system.

[0084] The dual-camera system consists of two closely connected industrial cameras and two targets. The main camera records the motion of the target deployed on the structure, and the sub-camera is used to measure the motion of the camera relative to the sub-target, which is located at a fixed reference point. Both the main target and the sub-target use the adaptive threshold-based gradient Hough circle transform (ATGHT) to extract displacement values. The extracted center coordinates of the deformed image are subtracted from the center coordinates in the reference image to obtain the pier settlement value during the construction of the main beam.

[0085] First, the Canny edge detection algorithm is used to detect the edges of the background-removed image output by the U 2 -net network, identify the circular edges, and output a binary image. In the binary image, the pixel gray value at the edge is 1, and the background gray value is 0. Then, the number of pixels with a gray value of 1 in the binary image N is counted; finally, the pixel radius value of the circle in the reference image and the deformed image is calculated. The radius value range for subsequent circle center detection is set to [0.9r, 1.1r].

[0086]

[0087] where r is the pixel radius value (pixel), n is the number of circular marker points to be detected, and N is the number of pixel points with a gray value of 1 in the binary image.

[0088] Then, a discrete feature curve f(r) is defined on the image gradient field to determine the radius of the circle in the image:

[0089]

[0090] where q(i,j) is the radial vector from the center to the pixel (i,j); f(r) is the average value of the dot product of the gradient vector and the radial vector; all pixels (i,j) satisfy the following formula:

[0091]

[0092] where Δr is the interval between adjacent r values; according to the feature definition of the gradient Hough circle transform, the non-zero gradient vector in the gradient field points to the center of the circle; the transform is defined using this feature to convert the gradient field into an accumulation array; the gradient intensity value of each pixel on the image edge determines the probability of the circle center, and the maximum intensity value in the accumulation image represents the center of the circular marker point. The center coordinates of the optical marker points at the main target and the sub-target are obtained;

[0093] Finally, the center coordinates of the deformed image are subtracted from the center coordinates of the reference image to obtain the displacement values at the main target and the sub-target; then, the measurement error of the camera self-motion is compensated, thereby obtaining the pier settlement values of the main beam during the cantilever construction process.

[0094] The dual-camera system uses the motion of the sub-camera to compensate for the measurement error caused by the camera self-motion, that is, the true displacement of the measured target is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, that is, the true settlement of the pier is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera.

[0095] The present application uses low-illumination image enhancement based on deep learning and a complex background segmentation algorithm to improve the image imaging quality and accurately segment the optical marker points; the gradient Hough transform based on adaptive threshold is adopted to extract the dynamic displacement of the multiple measurement points of the structure and combine the dual-camera system to compensate for the measurement error caused by the camera self-motion, so that the measurement result is more accurate.

[0096] Based on the above inventive concept, combined with Figures 2-6 , the present example further provides a non-contact monitoring system for pier settlement of a high-pier long-span continuous rigid frame bridge in a complex environment, and system component modules are as shown in Figure 7 , and include:

[0097] The image acquisition module 110 is used to acquire image sequences of the main target and the reference point sub-target of the pier under the construction condition of the main beam.

[0098] The image acquisition relies on a dual-camera visual monitoring system, and the visual imaging system refers to the hardware devices required for measurement, as shown in Figure 2 , Figure 3 The monitoring system adopts a dual-camera monitoring system, which is composed of two closely connected industrial cameras and two targets, and the two sets of industrial cameras (main camera and sub-camera) are placed on the same tripod and connected to a visual controller; optical marker points, i.e., main targets, are pasted at the pier settlement to be measured, and the motion of the main target disposed on the pier is monitored by the main camera of the dual-camera system; a reference point that is stationary is selected according to the site conditions, and optical marker points, i.e., sub-targets, are pasted thereon, and the camera self-motion is recorded by the sub-camera of the dual-camera system. At the same time, image sequences of the main target and the sub-target during the construction process of the main beam are collected, such as pouring of main beam segment concrete, main beam formwork hoisting, etc.

[0099] The image processing module 120 is used to process the images, including low-illumination image enhancement and high-precision segmentation of optical marker points in a complex background.

[0100] As shown in Figure 4As shown, there are low-illumination images and complex backgrounds in the image sequence, so the images need to be processed, and the image processing flow is as follows Figure 5 As shown, the SCI light self-correction model based on the training network of deep learning is used to perform light enhancement on the images, and the picture imaging quality is improved;The U 2 -net model is used to remove the complex background of the image, and the optical marker point is segmented with high precision.

[0101] The error compensation module 130: based on the adaptive threshold gradient Hough transform ATGHT, the optical marker point center coordinates of the main target and the sub-target are identified, the center coordinates of the deformed image are subtracted from the center coordinates in the reference image, and the displacement values of the main target and the reference point sub-target of the pier are obtained. Then, the camera self-motion measurement error is compensated (the motion of the sub-camera is used to compensate the measurement error caused by the camera self-motion in the double-camera system), that is, the real displacement of the pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, so as to obtain the pier settlement value of the main beam in the cantilever construction process, as shown in Figure 6 .

[0102] The present application uses a non-contact device to monitor the pier settlement displacement in the main beam construction stage, which can realize non-contact, multi-measuring point, high-precision and high-efficiency monitoring of the pier settlement displacement of the super-high pier long-span continuous rigid frame bridge under complex background, and has the advantages of low cost, convenient use and flexible measuring point selection.

[0103] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for monitoring the settlement displacement of a high-pier long-span continuous rigid frame bridge pier, characterized in that, It comprises the following steps: S1, collecting image sequences of the main target and the reference point sub-target at the bridge pier under the main girder construction condition; specifically comprising: S11, placing the dual-camera system on the same tripod and connecting it to the visual controller, wherein the dual-camera system comprises a main camera and a sub-camera; S12, pasting optical marker points, i.e. main targets, at the settlement of the bridge pier to be measured, and monitoring the movement of the main targets arranged on the bridge pier by the main camera of the dual-camera system; S13, selecting a stationary reference point according to the site conditions, pasting optical marker points, i.e. sub-targets, at the reference point, and recording the camera self-motion by the sub-camera of the dual-camera system; S14, simultaneously collecting image sequences of the main target and the sub-target during the construction of the main girder, including reference images and deformation images; S2, processing the images, including low-illumination image enhancement and high-precision segmentation of optical marker points in complex backgrounds; S3, identifying the optical marker point center coordinates at the main target and the sub-target based on the adaptive threshold gradient Hough transform ATGHT, further extracting the displacement values at the main target and the reference point sub-target of the bridge pier, and then compensating the camera self-motion measurement error, i.e. the real displacement of the bridge pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, so as to obtain the bridge pier settlement value during the cantilever construction of the main girder; specifically comprising: Firstly, the Canny edge detection algorithm is used to detect the edge of the background removed image output by the U 2 -net network, identify the circular edge and output a binary image; in the binary image, the pixel gray value at the edge is 1 and the background gray value is 0; the number N of pixels with a gray value of 1 in the binary image is counted, and the pixel radius value of the circle in the reference image and the deformed image is calculated, and the calculation formula is as follows: wherein r is the pixel radius value, n is the number of circular marker points to be detected, and N is the number of pixel points with a gray value of 1 in the binary image; Then, a discrete feature curve f(r) is defined on the image gradient field to determine the radius of the circle in the image: wherein q(i,j) is the radial vector from the center to the pixel (i,j); f(r) is the average value of the dot product of the gradient vector and the radial vector; the pixel (i,j) satisfies the following formula: wherein Δr is the interval between adjacent r values; According to the definition of the non-zero gradient vector pointing to the circular center in the gradient Hough circle transform ATGHT gradient field, the gradient field is converted into an accumulation array, the gradient intensity value of each pixel at the image edge determines the probability of the center of the circle, the maximum intensity value in the image represents the center of the circular marker point, and the center coordinates of the optical marker points at the main target and the sub-target are obtained; Finally, the center coordinates of the deformed image are subtracted from the center coordinates in the reference image to obtain the displacement values at the main target and the sub-target; then the camera self-motion measurement error is compensated, i.e. the real displacement of the bridge pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, to obtain the bridge pier settlement value during the construction of the main girder.

2. The method according to claim 1, wherein the method is characterized by: The low-illumination image enhancement in step S2 specifically improves the image imaging effect by using the SCI light self-calibration model to obtain a clear image; the SCI light self-calibration model introduces a weight-shared light learning and self-calibration module, and specifically comprises the following: Let the basic unit of the light optimization process be represented as: where u t and x t denote the residual term and illumination of the t-th stage, respectively, t = 0, 1, 2, …, T-1, H θ The illumination learning with weight sharing is used regardless of the stage number, and the illumination estimation network maintains the structure H and parameter θ shared in each stage; then the self-calibration module is introduced, so that the outputs of different stages in the training process can converge to the same state, and the self-calibration module is represented as: where v t is the input of the transition for each stage, is the introduced parameterized operator with learning parameters The basic unit of the illumination optimization process is transformed into:

3. The method according to claim 1, wherein the method is characterized by: The high-precision segmentation of the optical mark point in the complex background in the step S2 is specifically adopting a training network model U 2 -Net for segmenting the complex background of the image, removing the invalid background in the image, and only retaining the region of interest, and finally outputting the image after background removal by the U 2 -Net network; the U 2 -Net is a two-layer nested U-shaped structure and is composed of three parts: a 6-level encoder, a 5-level decoder, and an output fusion structure connected between the decoder and the last-level encoder. The encoder and decoder structure includes five residual network structures, respectively RSU-7, RSU-6, RSU-5, RSU-4 and RSU-4F, wherein RSU-4F uses dilated convolution to replace the up-sampling and down-sampling in the remaining structures; the output fusion structure fuses the saliency maps output by each layer to form the final prediction probability map.

4. The method according to claim 3, wherein the method is characterized by: The training network model U 2 The loss function Trainloss of the Net is the loss of the saliency map output by each layer of the network plus the loss of the prediction probability map of the final fusion output, and the calculation formula is: wherein, is the loss value of the output saliency map, l fuse is the loss value of the last output prediction probability map, and ω fuse is the weight coefficient of each part loss, and the standard binary cross entropy is used to calculate each loss value, and the calculation formula is: where (r, c) is the pixel coordinate, (H, W) is the image size, H represents the height, and W represents the width; P G(r,c) and P S(r,c) represent the pixel values of the Ground Truth and the predicted probability map, respectively; The model U 2 The training evaluation index F of Net β The calculation formulas of MAE and MSE are as follows: where β is a number between 0 and 1, and the code β is trained 2 is set to 0.3, F β The larger the value, the better the training effect. Wherein, P(r,c) is the prediction probability map, G(r,c) is the corresponding Ground Truth, MAE is a number between 0-1, and the smaller the value, the better the training effect.

5. The high pier and long span continuous rigid frame bridge pier settlement displacement monitoring system according to the high pier and long span continuous rigid frame bridge pier settlement displacement monitoring method of any one of claims 1-4, characterized in that: The system comprises: An image acquisition module (110) is configured to acquire image sequences of a main beam construction working condition of a bridge pier main target and a reference point sub-target; An image processing module (120) is configured to process the images, including low-illumination image enhancement and high-precision segmentation of optical marker points in a complex background; An error compensation module (130) is configured to identify the optical marker point center coordinates of the main target and the sub-target based on an adaptive threshold gradient Hough transform (ATGHT), further extract the displacement values of the bridge pier main target and the reference point sub-target, and then compensate for the camera self-motion measurement error, that is, the real displacement of the bridge pier settlement is equal to the structural displacement recorded by the main camera minus the camera self-motion recorded by the sub-camera, so as to obtain the bridge pier settlement value of the main beam in the cantilever construction process.

Images