Tunnel structure intelligent monitoring system based on machine vision

Abstract

The invention discloses a tunnel structure intelligent monitoring system based on machine vision, and relates to the technical field of tunnel monitoring. The method comprises the following steps: setting monitoring points and monitoring units in a tunnel, respectively acquiring a monitoring displacement value and a monitoring condition set of each monitoring point, acquiring point cloud data in the tunnel, constructing a three-dimensional model, respectively acquiring a calibration displacement value of each monitoring point in the three-dimensional model, and calculating the calibration displacement value of each monitoring point; constructing a displacement calibration model in the tunnel, obtaining a real-time calibration displacement value of each monitoring point and a change trend of the real-time calibration displacement value, constructing a structure evaluation model in the tunnel, obtaining abnormal structure types, generating corresponding abnormal structure signals according to different abnormal structure types, and feeding back the abnormal structure signals; the displacement value obtained through image analysis can be corrected in combination with point cloud data, and the abnormal structure existing in the tunnel can be judged in time according to the calibrated displacement value and change trend of each monitoring point and fed back.

Description

Technical Field

[0001] The invention relates to the technical field of tunnel monitoring, and in particular to an intelligent monitoring system for tunnel structures based on machine vision. Background Art

[0002] Intelligent monitoring of tunnel structures using machine vision is an advanced intelligent system that captures images and related data of tunnel structures through cameras and sensors installed inside the tunnel. It uses advanced image processing and data analysis algorithms to automatically identify and measure key parameters such as displacement and deformation of the tunnel, providing real-time monitoring and early warning for safe operation of the tunnel. However, the prior art has the following problems: although image analysis is highly efficient, it is subject to the parameter limitations of the camera, and the monitoring image will inevitably be distorted and deformed, resulting in obvious errors in the monitored displacement values. Although point cloud data can greatly improve the accuracy of data monitoring, its data processing volume is too large and it is difficult to achieve timely and effective feedback. If the two technologies can be combined, this problem can be solved. In view of the shortcomings of the prior art, the present invention provides a tunnel structure intelligent monitoring system based on machine vision. Summary of the invention

[0003] The purpose of the present invention is to provide a tunnel structure intelligent monitoring system based on machine vision.

[0004] The purpose of the present invention can be achieved through the following technical solutions: A tunnel structure intelligent monitoring system based on machine vision, comprising the following modules: The data acquisition module is used to set a number of monitoring points and different monitoring units inside the tunnel, and use the monitoring units to obtain corresponding monitoring data respectively, and obtain the monitoring displacement value and monitoring condition set of each monitoring point respectively according to the monitoring data; The data calibration module is used to obtain the point cloud data inside the tunnel and build the corresponding three-dimensional model, obtain the calibration displacement value of each monitoring point in the three-dimensional model, and build the displacement calibration model inside the tunnel according to the monitoring displacement values ​​of different monitoring points, the monitoring condition set and its calibration displacement value; The data evaluation module is used to obtain the real-time calibration displacement value using the displacement calibration model and obtain the corresponding change trend, build the structural evaluation model inside the tunnel in combination with the historical displacement data set, and obtain the corresponding abnormal structure type; The data feedback module is used to set the feedback unit to generate corresponding abnormal structure signals according to different abnormal structure types and provide feedback.

[0005] Furthermore, a process of setting a plurality of monitoring points and different monitoring units inside the tunnel and using the monitoring units to respectively obtain corresponding monitoring data includes: Acquire the supporting structures inside the tunnel, including the vault, side walls, and inverts, and set corresponding monitoring points at each supporting structure, on which passive targets are deployed; A monitoring unit is set up, including an image monitoring unit, a light monitoring unit, a vibration monitoring unit, a temperature monitoring unit, and a humidity monitoring unit, to respectively obtain corresponding monitoring data and its monitoring time, including monitoring images, light intensity, vibration amplitude, temperature, and humidity. The monitoring image contains all monitoring points within the monitoring range of the same image monitoring unit.

[0006] Furthermore, the process of respectively obtaining the monitoring displacement value and the monitoring condition set of each monitoring point according to the monitoring data includes: The first monitoring image acquired by the single image monitoring unit is used as its reference monitoring image, and each monitoring point in the reference monitoring image is used as a reference monitoring point, and other monitoring images other than the reference monitoring image acquired by the single image monitoring unit are used as its displacement monitoring images, and each monitoring point in the displacement monitoring image is used as a displacement monitoring point; The image features of the passive target at a single displacement monitoring point in a single displacement monitoring image are obtained by using an image processing algorithm, and the corresponding displacement monitoring coordinates are obtained; Obtain the reference monitoring coordinates of the reference monitoring point corresponding to the single displacement monitoring point, and obtain the horizontal pixel displacement value S of the single displacement monitoring point compared to the reference monitoring point according to the displacement monitoring coordinates and the reference monitoring coordinates. a and the vertical pixel displacement value S b ;

[0007]

[0008] Obtain the displacement monitoring coordinates of different displacement monitoring points in the single displacement monitoring image, and obtain the relative pixel displacement values ​​S between different displacement monitoring points c , (x 1 ,y 1 , z 1 ) is the reference monitoring coordinate, (x 2 ,y 2 , z 2 ) is the displacement monitoring coordinate, (x 3 ,y 3 , z 3 ) is another shift monitoring coordinate;

[0009] The pixel displacement value is converted into a physical displacement value, including a horizontal physical displacement value, a vertical physical displacement value, and a relative physical displacement value, by using an image processing algorithm combined with calibration parameters of an image monitoring unit; The horizontal physical displacement value, the vertical physical displacement value, and the relative physical displacement value are used as the monitoring displacement values ​​of the single displacement monitoring point at the corresponding monitoring time, and the light intensity, vibration amplitude, temperature, and humidity at the same monitoring time are used as the corresponding monitoring condition set.

[0010] Furthermore, the point cloud data inside the tunnel is obtained, and the corresponding three-dimensional model is constructed. The process of obtaining the calibration displacement value of each monitoring point in the three-dimensional model includes: A point cloud monitoring unit is provided. When the single image monitoring unit obtains its reference monitoring image, reference point cloud data inside the tunnel is obtained through the point cloud monitoring unit, and a three-dimensional reference model is constructed according to the reference point cloud data using three-dimensional modeling software; When the single image monitoring unit detects displacement of the monitoring point, it obtains the displacement point cloud data inside the tunnel and constructs a corresponding three-dimensional displacement model; Based on the position of the single image monitoring unit, the three-dimensional displacement model and the three-dimensional reference model are aligned to obtain a three-dimensional calibration model, and the calibration displacement value of the monitoring point where the displacement occurs is obtained. The calibration displacement value refers to each displacement value between two point cloud data corresponding to the monitoring point in the three-dimensional calibration model, including a horizontal calibration displacement value, a vertical calibration displacement value, and a relative calibration displacement value.

[0011] Furthermore, the process of constructing a displacement calibration model inside the tunnel according to the monitoring displacement values ​​of different monitoring points, the monitoring condition set and the calibration displacement values ​​thereof includes: Acquire a historical displacement data set of each monitoring point inside the tunnel where displacement has occurred, wherein the historical displacement data set includes a monitoring displacement value at a corresponding monitoring moment, a monitoring condition set and a calibration displacement value thereof, and divide the historical displacement data set into a first training set and a first test set; Constructing a first convolutional neural network, using different monitoring displacement values ​​and monitoring condition sets in the first training set as input data of the first convolutional neural network, and using corresponding calibration displacement values ​​in the first training set as output data of the first convolutional neural network; The first convolutional neural network is trained using the training set to obtain an initial first convolutional neural network, the initial first convolutional neural network is model verified using the test set, and an initial first convolutional neural network that is less than or equal to a preset first test error threshold is output as a displacement calibration model.

[0012] Furthermore, the process of obtaining the real-time calibration displacement value by using the displacement calibration model and obtaining the corresponding change trend includes: Inputting the real-time monitoring displacement value and monitoring condition set of a single monitoring point into the displacement calibration model to obtain corresponding real-time calibration displacement values, including real-time horizontal calibration displacement values, real-time vertical calibration displacement values, and real-time relative calibration displacement values; The real-time calibration displacement values ​​of each monitoring point within the monitoring range of the same image monitoring unit are obtained, and the monitoring points whose real-time horizontal calibration displacement values ​​are not equal to 0 are marked as horizontal change trends, the monitoring points whose real-time vertical calibration displacement values ​​are not equal to 0 are marked as vertical change trends, and the monitoring points whose real-time relative calibration displacement values ​​are not equal to 0 are marked as relative change trends.

[0013] Furthermore, the process of building a structural assessment model inside the tunnel in combination with the historical displacement data set and obtaining the corresponding abnormal structure type includes: The historical displacement data set includes the change trend of the calibrated displacement value of each monitoring point where displacement has occurred and its corresponding abnormal structure type, including convergence deformation, settlement deformation, and crack deformation; Generate a structure evaluation set according to the change trend of the calibration displacement value of each monitoring point and its abnormal structure type, and divide the obtained structure evaluation set into a second training set and a second test set; Constructing a second convolutional neural network, using different calibration displacement values ​​and change trends in the second training set as input data of the second convolutional neural network, and using the corresponding abnormal structure types in the second training set as output data of the second convolutional neural network; Using the training set to train the second convolutional neural network to obtain an initial second convolutional neural network, using the test set to perform model verification on the initial second convolutional neural network, and outputting an initial second convolutional neural network that is less than or equal to a preset second test error threshold as a structural evaluation model; The real-time calibrated displacement value of a single monitoring point and its variation trend are input into the structural assessment model to obtain the corresponding abnormal structure type.

[0014] Furthermore, a feedback unit is provided to generate corresponding abnormal structure signals according to different abnormal structure types and to provide feedback, and the process includes: A corresponding feedback unit is respectively set at each monitoring point to generate a convergence deformation signal for convergence deformation, a settlement deformation signal for settlement deformation, and a crack deformation signal for crack deformation; The abnormal structure signal includes a convergence deformation signal, a settlement deformation signal, and a crack deformation signal. The generated abnormal structure signal is fed back to relevant personnel and a corresponding alarm sound is issued.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention can obtain image data of each monitoring point and its corresponding environmental conditions by setting up multiple monitoring points and monitoring units inside the tunnel, and can judge and obtain the displacement of each monitoring point by analyzing the image data. By constructing a three-dimensional model at the corresponding monitoring time, the accurate displacement value of each monitoring point can be obtained, and the displacement value obtained by image analysis can be calibrated by using the three-dimensional model, and a corresponding displacement calibration model can be constructed. The displacement value obtained by image analysis can be corrected by combining the more accurate advantage of point cloud data. The corrected displacement value can be directly output by using the displacement calibration model, which can effectively give play to the faster advantage of image analysis and is conducive to the organic combination of the two technologies. Based on the historical displacement data sets of each monitoring point that has experienced displacement, it is possible to determine the deformation of the tunnel structure represented by the changing trends of different displacement values. By building a structural assessment model and combining it with a displacement calibration model, it is possible to efficiently determine what kind of abnormal structure exists based on the calibrated displacement values ​​and their changing trends of each monitoring point, and provide timely feedback to relevant personnel to prompt them to inspect and maintain the tunnel in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, a tunnel structure intelligent monitoring system based on machine vision includes the following modules: The data acquisition module is used to set a number of monitoring points and different monitoring units inside the tunnel, and use the monitoring units to obtain corresponding monitoring data respectively, and obtain the monitoring displacement value and monitoring condition set of each monitoring point respectively according to the monitoring data; The data calibration module is used to obtain the point cloud data inside the tunnel and build the corresponding three-dimensional model, obtain the calibration displacement value of each monitoring point in the three-dimensional model, and build the displacement calibration model inside the tunnel according to the monitoring displacement values ​​of different monitoring points, the monitoring condition set and its calibration displacement value; The data evaluation module is used to obtain the real-time calibration displacement value using the displacement calibration model and obtain the corresponding change trend, build the structural evaluation model inside the tunnel in combination with the historical displacement data set, and obtain the corresponding abnormal structure type; The data feedback module is used to set the feedback unit to generate corresponding abnormal structure signals according to different abnormal structure types and provide feedback.

[0018] It should be further explained that, in the specific implementation process, a number of monitoring points and different monitoring units are set inside the tunnel, and the process of using the monitoring units to obtain corresponding monitoring data includes: Obtain the supporting structure inside the tunnel, including the vault, side walls, and inverts. The vault refers to the structure at the top of the tunnel, used to support the overlying rock or soil layer to prevent collapse. The side walls refer to the structures on both sides of the tunnel, used to support the side walls and provide protection. The inverts refer to the structure at the bottom of the tunnel, used to support the foundation to prevent sinking. Corresponding monitoring points are respectively set at each supporting structure inside the tunnel, and passive targets are deployed on the monitoring points. The passive targets refer to markers pre-installed on the supporting structures inside the tunnel, specifically, highly reflective and high-contrast geometric patterns; Different monitoring units are arranged inside the tunnel, and the positions of the monitoring units are fixed by default, including an image monitoring unit, a light monitoring unit, a vibration monitoring unit, a temperature monitoring unit, and a humidity monitoring unit. The monitoring data inside the tunnel are obtained through the monitoring units, including monitoring images, light intensity, vibration amplitude, temperature, and humidity; The monitoring image contains all monitoring points within the monitoring range of the same image monitoring unit. The light intensity, vibration amplitude, temperature, and humidity refer to the environmental conditions when the monitoring image is acquired, and its timestamp is used as the corresponding monitoring moment.

[0019] It should be further explained that, in the specific implementation process, the process of obtaining the monitoring displacement value and the monitoring condition set of each monitoring point according to the monitoring data includes: The first monitoring image acquired by the single image monitoring unit is used as its reference monitoring image. Since the same monitoring image contains different monitoring points, each monitoring point in the reference monitoring image is used as a corresponding reference monitoring point; All monitoring images other than the reference monitoring image acquired by the single image monitoring unit are used as displacement monitoring images, and each monitoring point in the displacement monitoring image is used as a corresponding displacement monitoring point; The image features of the passive target at a single displacement monitoring point in a single displacement monitoring image are obtained by using an image processing algorithm, including the center coordinates, geometric shape, color, edge, and texture, and the center coordinates are used as the displacement monitoring coordinates; The same method is used to obtain the reference monitoring coordinates of the reference monitoring point corresponding to the single displacement monitoring point, and the horizontal pixel displacement value S of the single displacement monitoring point compared to the reference monitoring point is obtained according to the displacement monitoring coordinates and the reference monitoring coordinates. a and the vertical pixel displacement value S b ;

[0020]

[0021] Obtain the displacement monitoring coordinates of different displacement monitoring points in the single displacement monitoring image, and obtain the relative pixel displacement values ​​S between different displacement monitoring points c , the number of the relative pixel displacement values ​​is multiple;

[0022] Among them, (x 1 ,y 1 , z 1 ) is the reference monitoring coordinate, (x 2 ,y 2 , z 2 ) is the displacement monitoring coordinate, (x 3 ,y 3 , z 3 ) is another shift monitoring coordinate; The horizontal pixel displacement value, vertical pixel displacement value, and relative pixel displacement value are all pixel displacements of the passive target on the two-dimensional image. The pixel displacement value is converted into the physical displacement value of the tunnel structure in three-dimensional space by using the image processing algorithm combined with the calibration parameters of the image monitoring unit, including focal length, sensor size, and pixel size, including horizontal physical displacement value, vertical physical displacement value, and relative physical displacement value; The horizontal physical displacement value, the vertical physical displacement value, and the relative physical displacement value are used as the monitoring displacement values ​​of the single displacement monitoring point at the corresponding monitoring time, and the light intensity, vibration amplitude, temperature, and humidity at the same monitoring time are used as the corresponding monitoring condition set.

[0023] It should be further explained that, in the specific implementation process, the point cloud data inside the tunnel is obtained, and the corresponding three-dimensional model is constructed. The process of obtaining the calibration displacement value of each monitoring point in the three-dimensional model includes: A point cloud monitoring unit is set inside the tunnel. When the single image monitoring unit obtains its reference monitoring image, reference point cloud data inside the tunnel, including each monitoring point and monitoring unit, is obtained through the point cloud monitoring unit. A three-dimensional reference model inside the tunnel is constructed according to the obtained reference point cloud data using three-dimensional modeling software. The three-dimensional reference model refers to a three-dimensional model corresponding to the reference monitoring image. When the single image monitoring unit detects that there is displacement at the monitoring point, that is, at least one of the corresponding horizontal physical displacement value, vertical physical displacement value, and relative physical displacement value is not equal to 0, the displacement point cloud data inside the tunnel is obtained, and a corresponding three-dimensional displacement model is constructed, wherein the three-dimensional displacement model refers to a three-dimensional model corresponding to the displacement monitoring image at the corresponding monitoring time; Based on the position of the single image monitoring unit, the three-dimensional displacement model and the three-dimensional reference model are aligned to obtain a corresponding three-dimensional calibration model, wherein the three-dimensional calibration model includes all point cloud data of the two three-dimensional models, the positions of the single image monitoring unit overlap with each other, and the same monitoring point has two items of point cloud data in the three-dimensional calibration model; The calibration displacement value of the monitoring point where displacement occurs is obtained in the three-dimensional calibration model. The calibration displacement value refers to each displacement value between two point cloud data corresponding to the monitoring point in the three-dimensional calibration model, including horizontal calibration displacement value, vertical calibration displacement value, and relative calibration displacement value. The acquisition method thereof corresponds to the acquisition method of the monitoring displacement value.

[0024] It should be further explained that, in the specific implementation process, the process of constructing the displacement calibration model inside the tunnel according to the monitoring displacement values ​​of different monitoring points, the monitoring condition set and the calibration displacement values ​​thereof includes: Obtain a historical displacement data set of each monitoring point inside the tunnel where displacement has occurred, wherein the historical displacement data set includes a monitoring displacement value at a corresponding monitoring moment, a monitoring condition set and a calibration displacement value thereof, and divide the obtained historical displacement data set into a first training set and a first test set; Constructing a first convolutional neural network, using different monitoring displacement values ​​and monitoring condition sets in the first training set as input data of the first convolutional neural network, and using corresponding calibration displacement values ​​in the first training set as output data of the first convolutional neural network; The first convolutional neural network is trained using the training set to obtain an initial first convolutional neural network, the initial first convolutional neural network is model verified using the test set, and the initial first convolutional neural network that is less than or equal to a preset first test error threshold is output as the corresponding displacement calibration model.

[0025] It should be further explained that, in the specific implementation process, the process of obtaining the real-time calibration displacement value by using the displacement calibration model and obtaining the corresponding change trend includes: Inputting the real-time monitoring displacement value and monitoring condition set of a single monitoring point into the displacement calibration model to obtain corresponding real-time calibration displacement values, including real-time horizontal calibration displacement values, real-time vertical calibration displacement values, and real-time relative calibration displacement values; The same method is adopted to obtain the real-time calibration displacement value of each monitoring point within the monitoring range of the same image monitoring unit. If there is a monitoring point whose real-time horizontal calibration displacement value is not equal to 0, it is marked as a horizontal change trend. If there is a monitoring point whose real-time vertical calibration displacement value is not equal to 0, it is marked as a vertical change trend. If there is a monitoring point whose real-time relative calibration displacement value is not equal to 0, it is marked as a relative change trend.

[0026] It should be further explained that, in the specific implementation process, the process of building a structural assessment model inside the tunnel in combination with the historical displacement data set and obtaining the corresponding abnormal structure type includes: The historical displacement data set includes the change trend of the calibrated displacement value of each monitoring point where displacement has occurred and the corresponding abnormal structure type; The change trends include horizontal change trends, vertical change trends, and relative change trends, and the abnormal structural types include convergence deformation, settlement deformation, crack deformation, etc.; Generate a corresponding structure evaluation set according to the change trend of the calibration displacement value of each monitoring point and its abnormal structure type, and divide the obtained structure evaluation set into a second training set and a second test set; Constructing a second convolutional neural network, using different calibration displacement values ​​and change trends in the second training set as input data of the second convolutional neural network, and using the corresponding abnormal structure types in the second training set as output data of the second convolutional neural network; Using the training set to train the second convolutional neural network to obtain an initial second convolutional neural network, using the test set to perform model verification on the initial second convolutional neural network, and outputting an initial second convolutional neural network that is less than or equal to a preset second test error threshold as the corresponding structure evaluation model; The real-time calibrated displacement value of a single monitoring point and its change trend are input into the structural assessment model to obtain the corresponding abnormal structural type, including convergence deformation, settlement deformation, crack deformation, etc.

[0027] It should be further explained that, in the specific implementation process, the process of setting the feedback unit, generating corresponding abnormal structure signals according to different abnormal structure types and providing feedback includes: A corresponding feedback unit is respectively set at each monitoring point to generate a corresponding convergence deformation signal for convergence deformation, a corresponding settlement deformation signal for settlement deformation, and a corresponding crack deformation signal for crack deformation; The abnormal structure signal includes a convergence deformation signal, a settlement deformation signal, and a crack deformation signal. The generated abnormal structure signal is fed back to relevant personnel, and a corresponding alarm sound is issued to prompt relevant personnel to inspect and maintain the interior of the tunnel in a timely manner.

[0028] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A tunnel structure intelligent monitoring system based on machine vision, characterized in that: Includes the following modules: The data acquisition module is used to set a number of monitoring points and different monitoring units inside the tunnel, and use the monitoring units to obtain corresponding monitoring data respectively, and obtain the monitoring displacement value and monitoring condition set of each monitoring point respectively according to the monitoring data; The data calibration module is used to obtain the point cloud data inside the tunnel and build the corresponding three-dimensional model, obtain the calibration displacement value of each monitoring point in the three-dimensional model, and build the displacement calibration model inside the tunnel according to the monitoring displacement values ​​of different monitoring points, the monitoring condition set and its calibration displacement value; The data evaluation module is used to obtain the real-time calibration displacement value using the displacement calibration model and obtain the corresponding change trend, build the structural evaluation model inside the tunnel in combination with the historical displacement data set, and obtain the corresponding abnormal structure type; The data feedback module is used to set the feedback unit to generate corresponding abnormal structure signals according to different abnormal structure types and provide feedback.

2. The machine vision-based intelligent monitoring system for tunnel structures according to claim 1 is characterized in that: The process of setting up monitoring points and monitoring units and obtaining monitoring data includes: Acquire the supporting structures inside the tunnel, including the vault, side walls, and inverts, and set corresponding monitoring points at each supporting structure, on which passive targets are deployed; A monitoring unit is set up, including an image monitoring unit, a light monitoring unit, a vibration monitoring unit, a temperature monitoring unit, and a humidity monitoring unit, to respectively obtain corresponding monitoring data and its monitoring time, including monitoring images, light intensity, vibration amplitude, temperature, and humidity. The monitoring image contains all monitoring points within the monitoring range of the same image monitoring unit.

3. The machine vision-based intelligent monitoring system for tunnel structures according to claim 2 is characterized in that: The process of obtaining the monitoring displacement value and monitoring condition set of each monitoring point includes: The first monitoring image acquired by the single image monitoring unit is used as its reference monitoring image, and each monitoring point in the reference monitoring image is used as a reference monitoring point, and other monitoring images other than the reference monitoring image acquired by the single image monitoring unit are used as its displacement monitoring images, and each monitoring point in the displacement monitoring image is used as a displacement monitoring point; The image features of the passive target at a single displacement monitoring point in a single displacement monitoring image are obtained by using an image processing algorithm, and the corresponding displacement monitoring coordinates are obtained; Obtain the reference monitoring coordinates of the reference monitoring point corresponding to the single displacement monitoring point, and obtain the horizontal pixel displacement value S of the single displacement monitoring point compared to the reference monitoring point according to the displacement monitoring coordinates and the reference monitoring coordinates. a and the vertical pixel displacement value S b ; Obtain the displacement monitoring coordinates of different displacement monitoring points in the single displacement monitoring image, and obtain the relative pixel displacement values ​​S between different displacement monitoring points c , (x1, y1, z1) is the reference monitoring coordinate, (x2, y2, z2) is the displacement monitoring coordinate, and (x3, y3, z3) is another displacement monitoring coordinate; The pixel displacement value is converted into a physical displacement value, including a horizontal physical displacement value, a vertical physical displacement value, and a relative physical displacement value, by using an image processing algorithm combined with calibration parameters of an image monitoring unit; The horizontal physical displacement value, the vertical physical displacement value, and the relative physical displacement value are used as the monitoring displacement values ​​of the single displacement monitoring point at the corresponding monitoring time, and the light intensity, vibration amplitude, temperature, and humidity at the same monitoring time are used as the corresponding monitoring condition set.

4. The machine vision-based intelligent monitoring system for tunnel structures according to claim 3 is characterized in that: The process of obtaining the calibrated displacement values ​​of each monitoring point in the 3D model includes: A point cloud monitoring unit is provided. When the single image monitoring unit obtains its reference monitoring image, reference point cloud data inside the tunnel is obtained through the point cloud monitoring unit, and a three-dimensional reference model is constructed according to the reference point cloud data using three-dimensional modeling software; When the single image monitoring unit detects displacement of the monitoring point, it obtains the displacement point cloud data inside the tunnel and constructs a corresponding three-dimensional displacement model; Based on the position of the single image monitoring unit, the three-dimensional displacement model and the three-dimensional reference model are aligned to obtain a three-dimensional calibration model, and the calibration displacement value of the monitoring point where the displacement occurs is obtained. The calibration displacement value refers to each displacement value between two point cloud data corresponding to the monitoring point in the three-dimensional calibration model, including a horizontal calibration displacement value, a vertical calibration displacement value, and a relative calibration displacement value.

5. The machine vision-based intelligent monitoring system for tunnel structures according to claim 4 is characterized in that: The process of building a displacement calibration model of the tunnel interior includes: Acquire a historical displacement data set of each monitoring point inside the tunnel where displacement has occurred, wherein the historical displacement data set includes a monitoring displacement value at a corresponding monitoring moment, a monitoring condition set and a calibration displacement value thereof, and divide the historical displacement data set into a first training set and a first test set; Constructing a first convolutional neural network, using different monitoring displacement values ​​and monitoring condition sets in the first training set as input data of the first convolutional neural network, and using corresponding calibration displacement values ​​in the first training set as output data of the first convolutional neural network; The first convolutional neural network is trained using the training set to obtain an initial first convolutional neural network, the initial first convolutional neural network is model verified using the test set, and an initial first convolutional neural network that is less than or equal to a preset first test error threshold is output as a displacement calibration model.

6. The machine vision-based intelligent monitoring system for tunnel structures according to claim 5 is characterized in that: The process of obtaining the real-time calibration displacement value and its corresponding change trend includes: Inputting the real-time monitoring displacement value and monitoring condition set of a single monitoring point into the displacement calibration model to obtain corresponding real-time calibration displacement values, including real-time horizontal calibration displacement values, real-time vertical calibration displacement values, and real-time relative calibration displacement values; The real-time calibration displacement values ​​of each monitoring point within the monitoring range of the same image monitoring unit are obtained, and the monitoring points whose real-time horizontal calibration displacement values ​​are not equal to 0 are marked as horizontal change trends, the monitoring points whose real-time vertical calibration displacement values ​​are not equal to 0 are marked as vertical change trends, and the monitoring points whose real-time relative calibration displacement values ​​are not equal to 0 are marked as relative change trends.

7. The machine vision-based intelligent monitoring system for tunnel structures according to claim 6 is characterized in that: The process of building a structural assessment model inside the tunnel and obtaining the abnormal structure type includes: The historical displacement data set includes the change trend of the calibrated displacement value of each monitoring point where displacement has occurred and its corresponding abnormal structure type, including convergence deformation, settlement deformation, and crack deformation; Generate a structure evaluation set according to the change trend of the calibration displacement value of each monitoring point and its abnormal structure type, and divide the obtained structure evaluation set into a second training set and a second test set; Constructing a second convolutional neural network, using different calibration displacement values ​​and change trends in the second training set as input data of the second convolutional neural network, and using the corresponding abnormal structure types in the second training set as output data of the second convolutional neural network; Using the training set to train the second convolutional neural network to obtain an initial second convolutional neural network, using the test set to perform model verification on the initial second convolutional neural network, and outputting an initial second convolutional neural network that is less than or equal to a preset second test error threshold as a structural evaluation model; The real-time calibrated displacement value of a single monitoring point and its variation trend are input into the structural assessment model to obtain the corresponding abnormal structure type.

8. The machine vision-based intelligent monitoring system for tunnel structures according to claim 7 is characterized in that: The process of generating an abnormal structure signal and providing feedback according to the abnormal structure type includes: A corresponding feedback unit is respectively set at each monitoring point to generate a convergence deformation signal for convergence deformation, a settlement deformation signal for settlement deformation, and a crack deformation signal for crack deformation; The abnormal structure signal includes a convergence deformation signal, a settlement deformation signal, and a crack deformation signal. The generated abnormal structure signal is fed back to relevant personnel and a corresponding alarm sound is issued.

Images