A three-dimensional deformation monitoring and early warning method for reservoir dams based on binocular vision

Abstract

The present invention discloses a three-dimensional deformation monitoring and early warning method for a reservoir dam based on binocular vision. First, deformation feature points are arranged at key parts of the reservoir dam, and a method based on binocular video imaging is used to obtain high-precision continuous video sequence images. Then, the video sequence images are sequentially subjected to monocular camera feature point target tracking and matching, binocular camera feature point stereo matching, and coordinate conversion between the world coordinate system and pixel coordinates to obtain the three-dimensional spatial coordinates of each feature point. Finally, the deformation displacement of the feature point of the reservoir dam is quantified in real time based on the three-dimensional spatial coordinates of each feature point, so as to realize the three-dimensional deformation monitoring and early warning of the reservoir dam. The present invention breaks through the technical bottleneck that the original three-dimensional deformation of the reservoir dam is difficult to implement high-precision monitoring on a large scale, and can be integrated with the existing monitoring video equipment of the reservoir dam. It has the characteristics of low cost, easy installation, and convenient operation, and can meet the monitoring and early warning needs of real-time high-precision deformation of the reservoir dam.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent analysis of reservoir dams, and specifically relates to the design of a three-dimensional deformation monitoring and early warning method for reservoir dams based on binocular vision. Background Art

[0002] Reservoir dams are one of the important engineering measures for the development and utilization of water resources and the prevention and control of water disasters. They are crucial to flood control, water supply, ecology, power generation, and shipping. At present, my country has more than 98,000 reservoir dams, making it the country with the most reservoir dams in the world. More than 95% of them are earth-rock dams, and more than 95% are old dams built before the 1980s. At the same time, my country is also the country with the most dams above 200 meters in the world. At present, there are 77 dams above 200 meters built in the world, of which 20 are in my country, accounting for 26%; there are 19 dams above 200 meters under construction, of which 12 are in my country, accounting for 63%. Although my country has carried out large-scale reinforcement of dangerous reservoirs in the past 10 years, which has effectively improved the deformation safety status of reservoir dams, there are many reservoir dams with wide distribution, and most of them are earth-rock dams, which have a greater risk of deformation safety. In order to ensure the sustainable and safe operation of reservoir dams, it is urgent to carry out regular deformation safety monitoring of reservoir dams.

[0003] The means of monitoring and early warning of reservoir dam deformation are developing rapidly. The traditional method is to use a high-precision monitoring network to monitor the deformation elements of the reservoir dam. However, due to the influence of the terrain conditions of the reservoir dam, the network shape difference of the monitoring network and the position accuracy of the monitoring points are inaccurate, which affects the accuracy of the measurement. In addition, this method is very labor-intensive, takes a long time to observe, and does not achieve automated monitoring. With the emergence of global positioning navigation system (GNSS) technology, the deformation monitoring and early warning of reservoir dams have achieved continuous automation from data collection, data transmission, adjustment calculation and deformation analysis. The monitoring accuracy of horizontal displacement is less than ±2mm, and the measurement error of elevation is less than ±10mm. However, this method requires the installation of a ground base station, and there must be no communication obstruction above the deformation monitoring position. The equipment cost is high and the installation conditions are relatively harsh.

[0004] At present, video surveillance has been basically deployed in reservoir dams, and imaging data can be obtained in real time. However, there are still technical challenges in how to use binocular video imaging to calculate the deformation and displacement of specific parts of reservoir dams. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that it is difficult to implement high-precision monitoring of the three-dimensional deformation of existing reservoir dams on a large scale, and proposes a three-dimensional deformation monitoring and early warning method for reservoir dams based on binocular vision.

[0006] The technical solution of the present invention is: a three-dimensional deformation monitoring and early warning method for a reservoir dam based on binocular vision, comprising the following steps:

[0007] S1. Set up data acquisition equipment on the reservoir dam to obtain reference control points, feature points and video sequence images of the reservoir dam.

[0008] S2. Perform target tracking and matching of feature points of a monocular camera on the video sequence images of the reservoir dam to obtain the two-dimensional plane image coordinates of the feature points of each monocular video sequence image.

[0009] S3, performing stereo matching of the same-name stereo points of binocular camera feature points on the two-dimensional plane image coordinates of each monocular video sequence image feature point to obtain the same-name stereo matching point pair coordinates of the binocular video sequence image feature point.

[0010] S4, performing coordinate conversion on the stereo matching points of the same name of the feature points of the binocular video sequence images to obtain the three-dimensional space coordinates of each feature point at each moment.

[0011] S5. Compare the three-dimensional spatial coordinates of each feature point at each moment with the three-dimensional spatial coordinates of the reference control point at the same moment to obtain the deformation of each feature point at each moment.

[0012] S6. Conduct three-dimensional deformation monitoring and early warning of the reservoir dam based on the deformation of each feature point at each moment.

[0013] Furthermore, step S1 includes the following sub-steps:

[0014] S11. Use ground-based laser radar to obtain a three-dimensional model of the entire reservoir dam, conduct three-dimensional stress analysis based on the dam type and structure, and determine the key deformation areas of the reservoir dam based on the stress analysis results.

[0015] S12. At least 25 reference control points are evenly distributed in the key deformation area to establish a local control network. The three-dimensional spatial coordinates of the reference control points are obtained by high-precision total station or GNSS RTK with high-precision measurement, which are used for high-precision control of the binocular cameras set around the reservoir dam.

[0016] S13. 30 to 40 black and white circular artificial markers are evenly distributed at key positions in the key deformation area as feature points, and binocular cameras set around the reservoir dam are used to collect video sequence images of the reservoir dam in real time.

[0017] Further, step S2 includes the following sub-steps:

[0018] S21, for each monocular camera video sequence image, taking the first frame image as the reference, use the correlation coefficient R to perform rough matching in the search area to obtain the rough matching feature point coordinates (x 2 ,y 2 ):

[0019]

[0020] Among them I 1 (i,j) represents the gray value of the i-th row and j-th column of the first frame image, I 2 (i+x 2 ,j+y 2 ) represents the grayscale value of the image where the candidate feature points in the search area are located. Indicates I 1 The average gray value of (i,j), Indicates I 2 (i+x 2 ,j+y 2 )’s average gray value.

[0021] S22, according to the rough matching feature point coordinates (x 2 ,y 2 )Construct the error function E:

[0022] E=∑ i,j (I 1 (i,j)-I 2 (i+x 2 ,j+y 2 )) 2

[0023] S23, with the goal of minimizing the error function E, use the least squares method to accurately match the subsequent sequence images with the first frame image, and obtain the two-dimensional plane image coordinates of the feature points of each monocular video sequence image, that is, the two-dimensional plane image coordinates of the feature points of the left video sequence image and the two-dimensional plane image coordinates of the feature points of the right video sequence image

[0024] Further, step S3 includes the following sub-steps:

[0025] S31, based on the two-dimensional plane image coordinates of the left video sequence image feature points and the two-dimensional plane image coordinates of the feature points of the right video sequence image Calculate the normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera

[0026]

[0027]

[0028] Where K l represents the intrinsic parameter matrix of the left camera, K r Represents the intrinsic parameter matrix of the right camera.

[0029] S32, through the external parameter matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ] The normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera Convert to world coordinates:

[0030]

[0031]

[0032] in Represents the world coordinates of the left camera, R l represents the rotation matrix of the left camera, t l represents the translation vector of the left camera, Represents the world coordinates of the right camera, R r represents the rotation matrix of the right camera, t r Represents the translation vector of the right camera.

[0033] S33, according to the world coordinates of the left camera and the world coordinates of the right camera Determine the epipolar line of the feature points:

[0034]

[0035]

[0036] in represents the epipolar line direction vector, represents the intersection point of the epipolar lines.

[0037] S34, based on the geometric constraints of the kernel line, the two-dimensional search space is reduced to a one-dimensional kernel line space, and the coordinates of the stereo matching point pairs of the same name of the feature points of the binocular video sequence image are obtained. and

[0038] Further, step S4 includes the following sub-steps:

[0039] S41, based on the nonlinear model of the binocular camera under the influence of radial distortion, the distortion coefficient k of the binocular camera is calculated to be k=[k 1 k 2 ] T :

[0040] k=(D T D) -1 D T d

[0041]

[0042]

[0043] where k 1 and k 2 Both represent radial distortion parameters, D represents the first intermediate matrix, d represents the second intermediate matrix, (u, v) represents the pixel coordinates without distortion, represents the pixel coordinates with distortion, (u 0 ,v 0 ) represents the coordinates of the principal point, (x, y) represents the coordinates of the continuous image pixels without distortion, and the superscript T represents the transpose of the matrix.

[0044] S42, coordinates of stereo matching points with the same name according to the distortion coefficient k of the binocular camera and Perform distortion optimization to obtain the optimized coordinates of the stereo matching point pairs with the same name and

[0045]

[0046]

[0047] S43, according to the intrinsic parameter matrix K of the left camera l , the intrinsic parameter matrix K of the right camera r , the extrinsic matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ]Construct the projection matrix M of the left camera 1 and the projection matrix M of the right camera 2 :

[0048]

[0049]

[0050] in Represents the projection matrix M of the left camera 1 The elements in Represents the projection matrix M of the right camera 2 The elements in , a=1,2,3,4; b=1,2,3,4.

[0051] S44, according to the projection matrix M of the left camera 1 , the projection matrix M of the right camera 2 And the optimized coordinates of the stereo matching point pairs with the same name and Construct a camera imaging model and calculate the three-dimensional spatial coordinates of each feature point at each moment:

[0052]

[0053]

[0054]

[0055] Where (X i ,Y i ,Z i ) represents the three-dimensional space coordinates of each feature point at each moment.

[0056] Furthermore, the calculation formula for the deformation of each feature point at each moment in step S5 is:

[0057]

[0058]

[0059]

[0060] in represents the three-dimensional space coordinates of feature point i at time n, represents the three-dimensional space coordinates of the reference control point O at time n, (ΔX n ,ΔY n ,ΔZ n ) represents the deformation of feature point i at time n, and (Δx, Δy, Δz) represents the coordinate difference between feature point i and reference control point O at the initial time.

[0061] Further, step S6 includes the following sub-steps:

[0062] S61, according to the deformation (ΔX n ,ΔY n ,ΔZ n ) Calculate the horizontal displacement ΔH n and three-dimensional deformation displacement ΔR n :

[0063]

[0064]

[0065] S62, if the horizontal displacement ΔH n In the horizontal deformation monitoring early warning safety threshold [0-H max ], then go to step S63, otherwise a horizontal displacement deformation warning is issued.

[0066] S63, if the vertical displacement ΔZn In the vertical deformation monitoring early warning safety threshold [0-V max ], then go to step S64, otherwise a vertical displacement deformation warning is issued.

[0067] S64, if the three-dimensional deformation displacement ΔR n In the three-dimensional deformation monitoring early warning safety threshold [0-R max ], the feature point is safe, otherwise a 3D deformation warning is issued.

[0068] The beneficial effects of the present invention are:

[0069] (1) The present invention constructs a three-dimensional deformation monitoring and early warning method for reservoir dams based on binocular vision, breaking through the technical bottleneck of the original difficulty in implementing high-precision monitoring of three-dimensional deformation of reservoir dams on a large scale.

[0070] (2) The present invention adopts a binocular video imaging method to obtain the high-precision position coordinates of continuous reservoir dam deformation feature points, and performs stereo matching of the same name, thereby constructing a conversion relationship between the world coordinate system of the deformation feature points and the pixel coordinates, thereby realizing the high-precision three-dimensional coordinate acquisition of the reservoir dam deformation points and the quantification of deformation displacement.

[0071] (3) The present invention can be integrated with the existing monitoring video equipment of the reservoir dam, and has the characteristics of low cost, easy installation and convenient operation, and can meet the real-time and high-precision deformation monitoring and early warning needs of the reservoir dam. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 The figure shows a flow chart of a three-dimensional deformation monitoring and early warning method of a reservoir dam based on binocular vision provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0073] Now, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are only exemplary and are intended to explain the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0074] The embodiment of the present invention provides a three-dimensional deformation monitoring and early warning method for a reservoir dam based on binocular vision, such as Figure 1 As shown, the following steps S1 to S6 are included:

[0075] S1. Set up data acquisition equipment on the reservoir dam to obtain reference control points, feature points and video sequence images of the reservoir dam.

[0076] Step S1 includes the following sub-steps S11 to S13:

[0077] S11. Use ground-based laser radar to obtain a three-dimensional model of the entire reservoir dam, conduct three-dimensional stress analysis based on the dam type and structure, and determine the key deformation areas of the reservoir dam based on the stress analysis results.

[0078] S12. At least 25 reference control points are evenly distributed in the key deformation area to establish a local control network. The three-dimensional spatial coordinates of the reference control points are obtained by high-precision total station or GNSS RTK with high-precision measurement, which are used for high-precision control of the binocular cameras set around the reservoir dam.

[0079] S13. 30 to 40 black and white circular artificial markers are evenly arranged at key positions of the key deformation area (for example, a 5m×5m measurement area) as feature points, and a binocular camera set up around the reservoir dam is used to collect video sequence images of the reservoir dam in real time.

[0080] In the embodiment of the present invention, a CL600X2 binocular camera (resolution 1280pixel×1024pixel) is used. If the center extraction error of the mark point is 0.1pixel, the center positioning accuracy can reach 0.5mm (vertical direction: 5000mm / 1024pixel*0.1pixel=0.48mm, horizontal direction: 5000mm / 1280pixel*0.1pixel=0.39mm).

[0081] In the embodiment of the present invention, for night imaging or extreme weather conditions, artificial lighting facilities can be used to fill in the light so that the binocular camera can accurately capture the position of the artificial marker point.

[0082] S2. Perform target tracking and matching of feature points of a monocular camera on the video sequence images of the reservoir dam to obtain the two-dimensional plane image coordinates of the feature points of each monocular video sequence image.

[0083] In the embodiment of the present invention, for the sequential video image processing of each monocular camera, the target tracking of all feature points adopts a coarse-to-fine matching strategy, firstly using the correlation coefficient measurement to roughly match the search area, and then using the least squares matching method to accurately match the feature points.

[0084] Step S2 includes the following sub-steps S21 to S23:

[0085] S21, for each monocular camera video sequence image, taking the first frame image as the reference, use the correlation coefficient R to perform rough matching in the search area to obtain the rough matching feature point coordinates (x 2 ,y 2 ):

[0086]

[0087] Among them I 1 (i,j) represents the gray value of the i-th row and j-th column of the first frame image, I 2 (i+x 2 ,j+y 2 ) represents the grayscale value of the image where the candidate feature points in the search area are located. Indicates I 1 The average gray value of (i,j), Indicates I 2 (i+x 2 ,j+y 2 )’s average gray value.

[0088] S22, according to the rough matching feature point coordinates (x 2 ,y 2 )Construct the error function E:

[0089] E=∑ i,j (I 1 (i,j)-I 2 (i+x 2 ,j+y 2 )) 2

[0090] S23, with the goal of minimizing the error function E, use the least squares method to accurately match the subsequent sequence images with the first frame image, and obtain the two-dimensional plane image coordinates of the feature points of each monocular video sequence image, that is, the two-dimensional plane image coordinates of the feature points of the left video sequence image and the two-dimensional plane image coordinates of the feature points of the right video sequence image

[0091] S3, performing stereo matching of the same-name stereo points of binocular camera feature points on the two-dimensional plane image coordinates of each monocular video sequence image feature point to obtain the same-name stereo matching point pair coordinates of the binocular video sequence image feature point.

[0092] In the embodiment of the present invention, for binocular camera video sequence image processing, a stereo matching method with kernel line constraints is used to construct the homonymous relationship of feature points in the binocular camera sequence image to determine and obtain its homonymous points.

[0093] Step S3 includes the following sub-steps S31 to S34:

[0094] S31, based on the two-dimensional plane image coordinates of the left video sequence image feature points and the two-dimensional plane image coordinates of the feature points of the right video sequence image Calculate the normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera

[0095]

[0096]

[0097] Where K l represents the intrinsic parameter matrix of the left camera, K r Represents the intrinsic parameter matrix of the right camera.

[0098] S32, through the external parameter matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ] The normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera Convert to world coordinates:

[0099]

[0100]

[0101] in Represents the world coordinates of the left camera, R l represents the rotation matrix of the left camera, t l represents the translation vector of the left camera, Represents the world coordinates of the right camera, R r represents the rotation matrix of the right camera, t r Represents the translation vector of the right camera.

[0102] S33, according to the world coordinates of the left camera and the world coordinates of the right camera Determine the epipolar line of the feature points:

[0103]

[0104]

[0105] in represents the epipolar line direction vector, represents the intersection point of the epipolar lines.

[0106] S34, based on the geometric constraints of the kernel line, the two-dimensional search space is reduced to a one-dimensional kernel line space, and the coordinates of the stereo matching point pairs of the same name of the feature points of the binocular video sequence image are obtained. and

[0107] S4, performing coordinate conversion on the stereo matching points of the same name of the feature points of the binocular video sequence images to obtain the three-dimensional space coordinates of each feature point at each moment.

[0108] In the embodiment of the present invention, in order to obtain the accurate three-dimensional spatial coordinate accuracy of the dam feature points, it is first necessary to perform Zhang Zhengyou calibration method on the binocular camera to obtain high-precision camera internal and external parameters and distortion parameters, and then use the least squares method to solve the three-dimensional spatial coordinates of the feature points.

[0109] Step S4 includes the following sub-steps S41 to S44:

[0110] S41, based on the nonlinear model of the binocular camera under the influence of radial distortion, the distortion coefficient k of the binocular camera is calculated to be k=[k 1 k 2 ] T .

[0111] According to the imaging model of the camera, the relationship between the three-dimensional coordinate system and the plane coordinate system is constructed by shooting chessboard images with a binocular camera:

[0112]

[0113] Among them, S represents the scale factor, K represents the camera internal parameter, [R, T] represents the camera external parameter, R = [r 1 r 2 r 3 ] represents the rotation matrix, T represents the translation vector, (u,v) represents the pixel coordinates without distortion, and (X,Y,Z) represents the pixel three-dimensional space coordinates. Based on a set of checkerboard images taken at different angles, the camera intrinsic parameter K and the camera extrinsic parameters R and T can be solved and derived.

[0114] Due to the distortion in the left and right cameras of the binocular camera, the distortion will cause the image and the actual scene to not strictly satisfy the linear relationship. Therefore, the maximum likelihood estimation is used for optimization to build a nonlinear model that takes into account the influence of radial distortion:

[0115]

[0116] Rewrite it into matrix form:

[0117]

[0118] make The least square method is used to solve the distortion coefficient k = [k 1 k 2 ] T :

[0119] k=(D T D) -1 D T d

[0120] where k 1 and k 2Both represent radial distortion parameters, D represents the first intermediate matrix, d represents the second intermediate matrix, (u, v) represents the pixel coordinates without distortion, represents the pixel coordinates with distortion, (u 0 ,ν 0 ) represents the coordinates of the principal point, (x, y) represents the coordinates of the continuous image pixels without distortion, and the superscript T represents the transpose of the matrix.

[0121] S42, coordinates of stereo matching points with the same name according to the distortion coefficient k of the binocular camera and Perform distortion optimization to obtain the optimized coordinates of the stereo matching point pairs with the same name and

[0122]

[0123]

[0124] S43, according to the intrinsic parameter matrix K of the left camera l , the intrinsic parameter matrix K of the right camera r , the extrinsic matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ]Construct the projection matrix M of the left camera 1 and the projection matrix M of the right camera 2 :

[0125]

[0126]

[0127] in Represents the projection matrix M of the left camera 1 The elements in Represents the projection matrix M of the right camera 2 The elements in , a=1,2,3,4; b=1,2,3,4.

[0128] S44, according to the projection matrix M of the left camera 1 , the projection matrix M of the right camera 2 And the optimized coordinates of the stereo matching point pairs with the same name and Construct a camera imaging model and calculate the three-dimensional spatial coordinates of each feature point at each moment:

[0129]

[0130]

[0131]

[0132] Where (X i ,Y i ,Z i ) represents the three-dimensional space coordinates of each feature point at each moment.

[0133] S5. Compare the three-dimensional spatial coordinates of each feature point at each moment with the three-dimensional spatial coordinates of the reference control point at the same moment to obtain the deformation of each feature point at each moment.

[0134] Deformation refers to the change in shape and size of the characteristic points of the dam due to the action of external forces during the movement of the object. In the embodiment of the present invention, the calculation formula for the deformation of each characteristic point at each moment is:

[0135]

[0136]

[0137]

[0138] in represents the three-dimensional space coordinates of feature point i at time n, represents the three-dimensional space coordinates of the reference control point O at time n, (ΔX n ,ΔY n ,ΔZ n ) represents the deformation of feature point i at time n, and (Δx, Δy, Δz) represents the coordinate difference between feature point i and reference control point O at the initial time.

[0139] S6. Conduct three-dimensional deformation monitoring and early warning of the reservoir dam based on the deformation of each feature point at each moment.

[0140] Step S6 includes the following sub-steps S61 to S64:

[0141] S61, according to the deformation (ΔX n ,ΔY n ,ΔZ n ) Calculate the horizontal displacement ΔH n and three-dimensional deformation displacement ΔR n :

[0142]

[0143]

[0144] S62, if the horizontal displacement ΔH n In the horizontal deformation monitoring early warning safety threshold [0-H max], then go to step S63, otherwise a horizontal displacement deformation warning is issued.

[0145] S63, if the vertical displacement ΔZ n In the vertical deformation monitoring early warning safety threshold [0-V max ], then go to step S64, otherwise a vertical displacement deformation warning is issued.

[0146] S64, if the three-dimensional deformation displacement ΔR n In the three-dimensional deformation monitoring early warning safety threshold [0-R max ], the feature point is safe, otherwise a 3D deformation warning is issued.

[0147] In the embodiment of the present invention, for different reservoir dam types, such as concrete dam, earth-rock dam, slope and spillway, according to the reservoir dam position where the landmark is located, a high-precision total station is used to observe and obtain the maximum safety thresholds of monitoring and early warning in the horizontal, vertical and three-dimensional directions, and combined with the analysis results of the dam stress model, each threshold is optimized to obtain the deformation monitoring and early warning safety thresholds in each direction [0-H max ],[0-V max ] and [0-R max ].

[0148] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific deformations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A three-dimensional deformation monitoring and early warning method for reservoir dams based on binocular vision, characterized in that: The following steps are involved: S1. Set up data acquisition equipment on the reservoir dam to obtain reference control points, feature points and video sequence images of the reservoir dam; S2, performing target tracking and matching of feature points of a monocular camera on the video sequence images of the reservoir dam, and obtaining the two-dimensional plane image coordinates of the feature points of each monocular video sequence image; S3, performing stereo matching of the same-name stereo points of binocular camera feature points on the two-dimensional plane image coordinates of each monocular video sequence image feature point to obtain the same-name stereo matching point pair coordinates of the binocular video sequence image feature point; S4, performing coordinate conversion on the stereo matching points of the same name of the feature points of the binocular video sequence images to obtain the three-dimensional space coordinates of each feature point at each moment; S5, comparing the three-dimensional spatial coordinates of each feature point at each moment with the three-dimensional spatial coordinates of the reference control point at the same moment, to obtain the deformation of each feature point at each moment; S6. Conduct three-dimensional deformation monitoring and early warning of the reservoir dam based on the deformation of each feature point at each moment.

2. The three-dimensional deformation monitoring and early warning method of a reservoir dam based on binocular vision according to claim 1 is characterized in that: The step S1 comprises the following sub-steps: S11. Use ground-based laser radar to obtain a three-dimensional model of the entire reservoir dam, conduct three-dimensional stress analysis based on the dam type and structure, and determine the key deformation areas of the reservoir dam based on the stress analysis results; S12. At least 25 reference control points are evenly distributed in the key deformation area to establish a local control network. The three-dimensional spatial coordinates of the reference control points are obtained by high-precision measurement using a high-precision total station or GNSS RTK, which are used for high-precision control of the binocular cameras set around the reservoir dam; S13. 30 to 40 black and white circular artificial markers are evenly distributed at key positions in the key deformation area as feature points, and binocular cameras set around the reservoir dam are used to collect video sequence images of the reservoir dam in real time.

3. The three-dimensional deformation monitoring and early warning method of a reservoir dam based on binocular vision according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: S21. For each video sequence image of a monocular camera, taking the first frame image as a reference, a coarse matching is performed in the search area using the correlation coefficient R to obtain the coordinates of the coarse matching feature points (x2, y2): Where I1(i,j) represents the grayscale value of the i-th row and j-th column of the first frame image, and I2(i+x2,j+y2) represents the grayscale value of the image where the candidate feature point in the search area is located. represents the average gray value of I1(i,j), Represents the average gray value of I2(i+x2,j+y2); S22. Construct error function E according to the coordinates of the coarse matching feature points (x2, y2): E=Σ i,j (I1(i,j)-I2(i+x2,j+y2)) 2 S23, with the goal of minimizing the error function E, use the least squares method to accurately match the subsequent sequence images with the first frame image, and obtain the two-dimensional plane image coordinates of the feature points of each monocular video sequence image, that is, the two-dimensional plane image coordinates of the feature points of the left video sequence image and the two-dimensional plane image coordinates of the feature points of the right video sequence image 4. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on binocular vision according to claim 3 is characterized in that: The step S3 comprises the following sub-steps: S31, based on the two-dimensional plane image coordinates of the left video sequence image feature points and the two-dimensional plane image coordinates of the feature points of the right video sequence image Calculate the normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera Where K l represents the intrinsic parameter matrix of the left camera, K r Represents the intrinsic parameter matrix of the right camera; S32, through the external parameter matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ] The normalized camera coordinates of the left camera and the normalized camera coordinates in the right camera Convert to world coordinates: in Represents the world coordinates of the left camera, R l represents the rotation matrix of the left camera, t l represents the translation vector of the left camera, Represents the world coordinates of the right camera, R r represents the rotation matrix of the right camera, t r Represents the translation vector of the right camera; S33, according to the world coordinates of the left camera and the world coordinates of the right camera Determine the epipolar line of the feature points: in represents the epipolar line direction vector, represents the intersection of the epipolar lines; S34, based on the geometric constraints of the kernel line, the two-dimensional search space is reduced to a one-dimensional kernel line space, and the coordinates of the stereo matching point pairs of the same name of the feature points of the binocular video sequence image are obtained. and 5. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on binocular vision according to claim 4 is characterized in that: The step S4 comprises the following sub-steps: S41. Based on the nonlinear model of the binocular camera taking into account the influence of radial distortion, the distortion coefficient k of the binocular camera is calculated to be [k1 k2] T : k=(D T D) -1 D T d Where k1 and k2 both represent radial distortion parameters, D represents the first intermediate matrix, d represents the second intermediate matrix, (u, v) represents the pixel coordinates without distortion, represents pixel coordinates with distortion, (u0,v0) represents the coordinates of the principal point, (x,y) represents the continuous image pixel coordinates without distortion, and the superscript T represents the transpose of the matrix; S42, coordinates of stereo matching points with the same name according to the distortion coefficient k of the binocular camera and Perform distortion optimization to obtain the optimized coordinates of the stereo matching point pairs with the same name and S43, according to the intrinsic parameter matrix K of the left camera l , the intrinsic parameter matrix K of the right camera r , the extrinsic matrix of the left camera [R l |t l ] and the external parameter matrix of the right camera [R r |t r ]Construct the projection matrix M1 of the left camera and the projection matrix M2 of the right camera: in represents the element in the projection matrix M1 of the left camera, Represents the elements in the projection matrix M2 of the right camera, a=1,2,3,4; b=1,2,3,4; S44, according to the projection matrix M1 of the left camera, the projection matrix M2 of the right camera and the optimized coordinates of the stereo matching points with the same name and Construct a camera imaging model and calculate the three-dimensional spatial coordinates of each feature point at each moment: Where (X i ,T i ,Z i ) represents the three-dimensional space coordinates of each feature point at each moment.

6. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on binocular vision according to claim 5 is characterized in that: The calculation formula of the deformation of each feature point at each moment in step S5 is: in represents the three-dimensional space coordinates of feature point i at time n, represents the three-dimensional space coordinates of the reference control point O at time n, (ΔX n ,ΔY n ,ΔZ n ) represents the deformation of feature point i at time n, and (Δx, Δy, Δz) represents the coordinate difference between feature point i and reference control point O at the initial time.

7. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on binocular vision according to claim 6 is characterized in that: The step S6 comprises the following sub-steps: S61, according to the deformation (ΔX n ,ΔY n ,ΔZ n ) Calculate the horizontal displacement ΔH n and three-dimensional deformation displacement ΔR n : S62, if the horizontal displacement ΔH n In the horizontal deformation monitoring early warning safety threshold [0-H max ], then go to step S63, otherwise a horizontal displacement deformation warning is performed; S63, if the vertical displacement ΔZ n In the vertical deformation monitoring early warning safety threshold [0-V max ], then go to step S64, otherwise a vertical displacement deformation warning is performed; S64, if the three-dimensional deformation displacement ΔR n In the three-dimensional deformation monitoring early warning safety threshold [0-R max ], the feature point is safe, otherwise a 3D deformation warning is issued.

Images