Water stop needle remote installation system and method for underwater crack repair of tunnels

Abstract

The application discloses a water-stopping needle remote installation system and method for underwater crack repair of a tunnel, which comprises a master end, a communication channel and an underwater robot of a slave end, the master end is connected with the underwater robot of the slave end through the communication channel, communicates with the underwater robot of the slave end, controls the slave end to work, the master end at least comprises a force feedback hand controller and a remote control terminal, the remote control terminal restores underwater images based on a cycle generative adversarial network (CycleGAN) according to left and right eye images of a binocular camera, completes hole position detection by combining a deep learning method Yolov5, obtains a disparity map by combining a binocular stereo matching method SGBM, completes three-dimensional point cloud reconstruction, fits hole position point cloud information by using a RANSAC algorithm, adjusts a mechanical arm end to be perpendicular to the hole position according to hole position pose information, and installs the water-stopping needle, so that the automation level of a remote installation operation of the water-stopping needle for the crack of the tunnel concrete lining is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of underwater robot detection and operation, and particularly relates to a water stop needle remote installation system and method for underwater crack repair of a tunnel. BACKGROUND

[0002] Water conveyance tunnels are important infrastructure for water resource utilization in China. The tunnels realize water supply function and guarantee human production and living needs. With long-time use of the tunnels, defects such as corrosion and leakage of concrete lining occur, which seriously threatens the water supply capacity of the tunnels. Therefore, the tunnels need to be regularly overhauled and maintained. On the basis of defect positioning, a method for quickly repairing the concrete lining is developed to guarantee the safety of water conveyance.

[0003] After the drilling task near the lining crack is completed, high-pressure grouting needs to be performed around the lining crack to prevent water leakage, deformation or collapse of the tunnel. The water stop needle is a professional component for high-pressure chemical grouting plugging construction and is widely applied in the fields of water conservancy and hydropower, tunnels, industry and construction. The water stop needle is installed in the hole near the lining crack to complete the high-pressure grouting operation. Due to low visibility and long distance of the underwater environment of the water conveyance tunnel, manual construction operation is difficult and has great safety hazards. These factors seriously restrict the manual operation method and reduce the operation efficiency.

[0004] Therefore, it is urgent to develop a water stop needle remote installation method for underwater crack repair of a water conservancy tunnel to quickly position the drilling position and complete the water stop needle assembly work, thereby improving the operation efficiency. SUMMARY

[0005] The present application is aimed at the problems of difficult and slow positioning of drilling position and low work efficiency of water stop needle installation in the prior art, and provides a water stop needle remote installation system and method for underwater crack repair of a tunnel, which comprises a master end, a communication channel and a slave end underwater robot, the master end is connected to the slave end underwater robot through the communication channel and communicates with the slave end underwater robot to control the slave end to work, the slave end underwater robot comprises at least a monocular camera and left and right binocular cameras, the monocular camera is located at the center of the top of the underwater robot, and the binocular cameras are located at the ends of the mechanical arms of the underwater robot, the master end comprises at least a force feedback hand controller and a remote control terminal, the remote control terminal restores underwater images based on a cycle generative adversarial network (CycleGAN) according to left and right eye images of the binocular camera to obtain clear underwater images by inputting original underwater images into a network structure, underwater dataset pictures are collected in advance, the underwater dataset pictures after clarity restoration are manually labeled, a hole position area is labeled on the pictures with a label frame, and a dataset is trained based on a deep learning method Yolov5 to obtain a trained model, when a left eye picture after CycleGAN network restoration is obtained, the trained Yolov5 network structure is inputted to obtain a hole position anchor frame and complete hole position detection, a disparity map is obtained by combining a binocular stereo matching method SGBM, holes in the disparity map are compensated, three-dimensional point cloud reconstruction is completed, RANSAC algorithm is used to fit hole position point cloud information, the end of the mechanical arm is adjusted to be perpendicular to the hole position according to hole position pose information, and the water stop needle is installed, thereby improving the automation level of remote installation operation of the water stop needle for cracks in the tunnel concrete lining.

[0006] To achieve the above object, the technical scheme adopted by the present application is as follows: a water stop needle remote installation system for underwater crack repair of a tunnel, which comprises a master end, a communication channel and a slave end underwater robot, the master end is connected to the slave end underwater robot through the communication channel and communicates with the slave end underwater robot to control the slave end to work,

[0007] The slave end underwater robot comprises at least a monocular camera and left and right binocular cameras, the monocular camera is located at the center of the top of the underwater robot, and the binocular cameras are located at the ends of the mechanical arms of the underwater robot;

[0008] The master end at least includes a force feedback hand controller and a remote control terminal, the remote control terminal carries out underwater image restoration based on a cycle generation adversarial network CycleGAN according to left and right eye images of a binocular camera, inputs an original underwater image into a network structure to obtain a clear underwater image, pre-collects underwater dataset pictures, manually labels the underwater dataset pictures after clarity restoration, labels a hole position area on the pictures with a label frame, trains the dataset based on a deep learning method Yolov5 to obtain a trained model, when the left eye picture after CycleGAN network restoration is obtained, inputs the already trained Yolov5 network structure to obtain a hole position anchor frame, at this time, hole position detection is completed, a disparity map is obtained by combining a binocular stereo matching method SGBM, holes in the disparity map are compensated, three-dimensional point cloud reconstruction is completed, outliers are removed by using Gaussian filtering, hole position point cloud information is fitted by using an RANSAC algorithm, a hole position plane is calculated, hole center point coordinates and a normal vector are obtained, according to hole position pose information, the end of the mechanical arm is adjusted to be perpendicular to the hole position, and installation of a water stop needle is carried out.

[0009] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is also: a water stop needle remote installation method for underwater crack repair of a tunnel, comprising the following steps:

[0010] S1: video information of an underwater environment is fed back to a remote control terminal and displayed by a slave monocular camera, an operator observes underwater cracks and hole positions, and determines a preliminary insertion hole position of a water stop needle;

[0011] S2: the operator remotely controls the end of an underwater mechanical arm to approach the preliminarily selected hole position by using a force feedback hand controller;

[0012] S3: according to left and right eye images of a binocular camera arranged at the end of the mechanical arm, underwater image restoration is carried out based on a cycle generation adversarial network CycleGAN, an original underwater image is input into a network structure to obtain a clear underwater image, underwater dataset pictures are pre-collected, underwater dataset pictures after clarity restoration are manually labeled, a hole position area is labeled on the pictures with a label frame, and a dataset is trained based on a deep learning method Yolov5 to obtain a trained model, when a left eye picture after CycleGAN network restoration is obtained, the already trained Yolov5 network structure is input to obtain a hole position anchor frame, and hole position detection is completed;

[0013] S4: a disparity map is obtained by combining a binocular stereo matching method SGBM, holes in the disparity map are compensated, and three-dimensional point cloud reconstruction is completed;

[0014] S5: three-dimensional point clouds obtained in step S4 are transmitted to a remote control terminal through a communication channel and displayed on a visual interface, Gaussian filtering is performed on global point clouds to remove outliers, and at the same time, hole position point clouds are segmented according to hole position information obtained in step S3.

[0015] S6: fitting the hole site cloud information by using the RANSAC algorithm, and obtaining the plane equation H of the hole according to the point cloud information:

[0016] H:n x x+n y y+n z z+d=0

[0017] Wherein, (n x ,n y ,n z ) is the component of the normal vector, and d is the distance from the origin to the plane; according to the plane equation H and the point cloud result, the current hole center coordinates (o x ,o y ,o z ) and the normal vector (n x ,n y ,n z ) are calculated.

[0018] S7: according to the D-H parameter method, the homogeneous transformation matrix between the adjacent links of the mechanical arm is solved, the homogeneous transformation matrix between the links of the mechanical arm is multiplied, and the pose equation of the end effector relative to the base is obtained. According to the hole position information obtained in step S6, the inverse kinematics of the mechanical arm is solved, the rotation angles θ1, θ2, θ3, θ4 of each joint of the mechanical arm are calculated, and the pose of the end effector of the mechanical arm is adjusted to be perpendicular to the hole position.

[0019] S8: combining the compliance control to complete the compliant installation of the water stop needle insertion hole, according to the six-dimensional force sensor, the pose of the end effector of the mechanical arm is constantly adjusted, and the blocking phenomenon is eliminated.

[0020] S9: when the six-dimensional force sensor at the end of the mechanical arm measures that the assembly direction z-axis force is greater than a specified threshold, and the hole depth and the insertion water stop needle depth are less than a specified threshold, the water stop needle installation is successful, and the step is ended; otherwise, steps S7-S8 are continued.

[0021] Compared with the prior art, the present application has the following technical effects and advantages:

[0022] 1. The present application proposes a water stop needle remote installation method and system suitable for underwater tunnel environment, which can quickly complete the tunnel concrete lining crack plugging work, improve the automation level and safety of assembly, and significantly improve the underwater robot operation efficiency.

[0023] 2. The system of the present application is based on a teleoperation system, which is divided into three parts of master, communication channel and underwater robot, displays the working environment information in the master visual interface, and assists the operator to move the end of the mechanical arm to the vicinity of the hole position through visual information feedback, thereby improving the accuracy of subsequent assembly.

[0024] 3.The application is based on binocular vision method to realize automatic positioning, based on cycle generative adversarial network CycleGAN to restore underwater image and combined with deep learning method Yolov5 to complete hole detection; combined with binocular stereo matching method to get disparity map, complete three-dimensional point cloud reconstruction; according to the hole area information, based on RANSAC method to complete the accurate segmentation of hole point cloud and calculate the hole pose information, improve the detection speed and the accuracy of pose information.

[0025] 4.The installation stage of the water stop needle insertion hole of the application is based on admittance control to complete the compliant assembly, which combines the information of the six-dimensional force sensor at the end of the mechanical arm, continuously adjusts the pose of the end effector of the mechanical arm, and is beneficial to eliminate the blocking phenomenon in the installation process. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 It is a structural schematic diagram of the water stop needle remote installation system for underwater crack repair of a tunnel of the application;

[0027] Figure 2 It is a structural schematic diagram of the slave robot in the water stop needle remote installation system for underwater crack repair of a tunnel of the application;

[0028] Figure 3 It is a step flowchart of the water stop needle remote installation method for underwater crack repair of a tunnel of the application;

[0029] Figure 4 It is a flowchart of the automatic positioning stage in the water stop needle remote installation method for underwater crack repair of a tunnel of the application;

[0030] Figure 5 It is a point cloud reconstruction schematic diagram based on the binocular vision method in step S4 of the water stop needle remote installation method for underwater crack repair of a tunnel of the application;

[0031] Figure 6 It is a water stop needle insertion hole installation control diagram based on admittance control in step S8 of the method of the application;

[0032] LIST OF FIGURES:

[0033] 1-1, operator; 1-2, force feedback hand controller; 1-3, remote control terminal; 1-4, photoelectric composite cable, 1-5, tunnel; 1-6, underwater robot; 1-7, crack and hole;

[0034] 2-1, walking wheel; 2-2, track; 2-3: chassis; 2-4: searchlight; 2-5, telescopic rod; 2-6, monocular camera; 2-7, mechanical arm;

[0035] 3-1, base; 3-2, six-dimensional force sensor; 3-3, binocular camera; 3-4, water stop needle;

[0036] 4-1, segmented hole site three-dimensional point cloud model; 4-2, crack point cloud model. DETAILED DESCRIPTION

[0037] The present application will be further illustrated in conjunction with the accompanying drawings and specific embodiments, and it should be understood that the following specific embodiments are only used to illustrate the present application and not to limit the scope of the present application.

[0038] Embodiment 1

[0039] The water stop needle remote installation system for underwater crack repair of the tunnel is shown in Figure 1 It is mainly used for crack and hole repair operation in the tunnel 1-5, and includes a master end, a communication channel and a slave end underwater robot 1-6. The master end is connected to the slave end underwater robot 1-6 through the communication channel and communicates with the slave end underwater robot 1-6 to control the slave end work.

[0040] As shown in Figure 1 and Figure 3 The master end at least includes a force feedback hand controller 1-2 and a remote control terminal 1-3. The communication channel realizes long-distance wired communication through a remote control optical and electrical composite cable 1-4. The slave end underwater robot 1-6 at least includes a monocular camera 2-6 and left and right binocular cameras 3-3. The monocular camera 2-6 is located at the center of the top of the underwater robot 1-6. The monocular camera 2-6 is divided into a telescopic rod 2-5 with a holder and a monocular camera. The holder can adjust the height and rotation angle of the camera to adjust the field of view and help the operator better determine the water stop needle insertion hole position. The binocular camera 3-3 is located at the end of the mechanical arm 2-7 of the underwater robot 1-6. The binocular camera 3-3 is erected at the end of the mechanical arm 2-7 through a mounting mechanism. The mounting mechanism can stably fix the binocular camera 3-3 to prevent shaking. A waterproof protective shell is provided outside the binocular camera 3-3 to ensure stable operation.

[0041] The underwater robot 1-6 in the embodiment further includes walking wheels 2-1, tracks 2-2, a chassis 2-3, and a track chassis for underwater movement of the robot. A plurality of searchlights 2-4 are installed on the track vehicle. The searchlights 2-4 are installed at the front end of the chassis 2-3 and are used to illuminate the dark underwater environment, facilitating the operator 1-1 to observe the underwater crack and hole position. The robot mechanical arm 2-7 in the embodiment is installed on the chassis 2-3 through the base 3-1. The mechanical arm 2-7 is designed with waterproof sealing for water stop needle 3-4 installation operation. The mechanical arm 2-7 is provided with a force sensor 3-2 at the end and the end mechanism is adapted to the water stop needle 3-4 installation.

[0042] In the system, the monocular camera 2-6 mounted on the underwater robot 1-6 feeds back the video information of the underwater environment to the remote control terminal 1-3 and displays, the operator 1-1 observes the underwater crack and hole 1-7, and preliminarily determines the insertion hole position of the water stop needle head 3-4. The spatial position of the force feedback hand controller 1-2 at the end is mapped to the end of the mechanical arm 2-7 by using a position mapping mechanism. The operator 1-1 controls the force feedback hand controller 1-2 to remotely control the end of the underwater mechanical arm 2-7 to approach the selected hole position, so as to facilitate the autonomous positioning of the binocular camera 3-3 later.

[0043] Subsequently, for the left and right eye images of the binocular camera 3-3 erected at the end of the mechanical arm 2-7, the CycleGAN image processing algorithm is applied to the left and right eye images of the binocular camera 3-3 erected at the end of the mechanical arm 2-7, and the original underwater image is input into the network structure to obtain a clear underwater image. The image processing process weakens the interference of the underwater environment on the image and improves the clarity of the image. The underwater dataset pictures are collected in advance, the underwater dataset pictures after clarity recovery are manually labeled, the hole position area is labeled on the pictures with a label frame, and the dataset is trained based on the deep learning method Yolov5 to obtain the trained model. When the left eye picture after CycleGAN network recovery is obtained, the Yolov5 network structure which has been trained is input to obtain the hole position anchor frame, and the hole position detection is completed. The method has short training time and is robust for hole position detection.

[0044] The parallax map is obtained by combining the binocular stereo matching method SGBM, the parallax map holes are compensated, the three-dimensional point cloud reconstruction is completed, the global point cloud is subjected to Gaussian filtering to remove outliers, and the hole position point cloud is segmented according to the hole position information. The point cloud information is fitted by using the RANSAC algorithm, the plane equation H of the hole is calculated, the center point coordinate (o x y z ) and the normal vector (n x y z ) of the current hole position are calculated. According to the D-H parameter method, the homogeneous transformation matrix between the adjacent links of the mechanical arm 2-7 is solved, the homogeneous transformation matrices between the links of the mechanical arm 2-7 are multiplied, and the pose equation of the end effector relative to the base is obtained. According to the obtained hole position pose information, the inverse kinematics of the mechanical arm 2-7 is solved, the rotation angles θ1, θ2, θ3, θ4 of the joints of the mechanical arm 2-7 are calculated, and the pose of the end effector of the mechanical arm 2-7 is adjusted to be perpendicular to the hole position.

[0045] ​​​​The six-dimensional force sensor 3-2 is installed at the end of the mechanical arm 2-7. When the mechanical arm 2-7 end feels external force, it is determined that the hole and the water stop needle 3-4 are in contact. The compliant assembly method of the water stop needle 3-4 insertion hole is completed by combining the admittance control. The joint position of the mechanical arm 2-7 end is controlled. The mechanical arm 2-7 end effector pose is constantly fine-tuned to eliminate the blocking phenomenon, so that the water stop needle 3-4 and the hole can be smoothly assembled, and the installation of the water stop needle 3-4 is completed.

[0046] When the six-dimensional force sensor 3-2 at the end of the mechanical arm 2-7 measures that the assembly direction z-axis force is greater than the specified threshold, and the hole depth and the insertion depth of the water stop needle 3-4 are less than the specified threshold, it is considered that the water stop needle 3-4 insertion hole assembly is successful, and the water stop needle 3-4 installation work is completed.

[0047] Embodiment 2

[0048] The water stop needle remote installation method for underwater crack repair of a tunnel can be divided into three stages. The first stage is a manual positioning stage. The second stage is an automatic positioning stage. The third stage is a water stop needle insertion hole installation stage, which includes the following steps: Figure 3

[0049] Step S1: Enter the manual positioning stage. The monocular camera 2-6 installed on the underwater robot 1-6 feeds back the underwater environment video information to the remote control terminal 1-3. The operator 1-1 observes the underwater crack and the hole 1-7, and preliminarily determines the position of the water stop needle 3-4 insertion hole.

[0050] Step S2: The operator 1-1 controls the force feedback hand controller 1-2 to remotely control the underwater mechanical arm 2-7 end to approach the hole position selected in step S1, so as to facilitate subsequent autonomous positioning of the binocular camera 3-3. The force feedback hand controller 1-2 has 3 position degrees of freedom, 3 joint degrees of freedom and 3 degree of freedom force feedback output. A position mapping mechanism is adopted to map the spatial position of the end of the force feedback hand controller 1-2 to the end of the mechanical arm 2-7. The mapping relationship is:

[0051]

[0052] Where [x p ,y p ,z p ] is the position of the end of the force feedback hand controller, [x, y, z] is the position of the end of the underwater mechanical arm, a x , a y , a z are the magnification multiples along the x-axis, y-axis and z-axis directions, respectively.

[0053] Step S3: Enter the automatic positioning stage, as shown in Figure 4 ​As shown, according to the left and right eye images of the binocular camera 3-3 mounted at the end of the mechanical arm 2-7, the underwater image restoration is performed based on the cycle generation adversarial network CycleGAN, the original underwater image is input into the network structure to obtain a clear underwater image, this method can improve the clarity of the underwater image and improve the recognition accuracy. The underwater dataset pictures are collected in advance, the underwater dataset pictures after clarity restoration are manually labeled, the hole position area is labeled on the pictures with a label box, and the dataset is trained based on the deep learning method Yolov5 to obtain the trained model. When the left eye picture after CycleGAN network restoration is obtained, the Yolov5 network structure which has been trained is input to obtain the hole position anchor frame and complete the hole position detection.

[0054] Step S4: obtain the parallax map by combining the binocular stereo matching method SGBM, compensate the parallax map holes, complete the three-dimensional point cloud reconstruction, and transmit the three-dimensional point cloud to the remote control terminal 1-3 through the communication channel and display it on the visualization interface 1-3, as shown in Figure 5 As shown, the point cloud data is recorded in the form of points, containing spatial three-dimensional coordinate information, which is convenient for subsequent calculation of the hole position information. Figure 5 4-1 in is the segmented hole position three-dimensional point cloud model, and 4-2 is the crack point cloud model.

[0055] Step S5: Gaussian filtering is performed on the global point cloud to remove outliers, solving the problems of holes, noise data and the like in the point cloud data. Based on KD-Tree, the k neighborhood points around each point cloud data are searched in turn, and the average Euclidean distance from the sampling point to the k neighborhood points is calculated. Let p i ={x i ,x i ,x i |i=1,2,3,…,n} be the original point cloud data. After KD-Tree search, the data set is p j ={x j ,x j ,x j |j=1,2,3,…,k}., then the following formula is obtained:

[0056]

[0057]

[0058]

[0059] where d i is the average distance from the point p j to its k neighborhood points, is the mean value of d , and sigma is d istandard deviation. We give its mean and variance, remove the point cloud outliers. According to the hole site information obtained in step S3, the hole site point cloud is segmented.

[0060] Step S6: fitting hole site point cloud information using RANSAC algorithm. k is the maximum number of iterations, p is the confidence, w is the proportion of inliers, and m is the minimum number of samples required for the calculation model:

[0061]

[0062] According to the point cloud information, the plane equation H of the hole is obtained:

[0063] H:n x x+n y y+n z z+d=0

[0064] where (n x ,n y ,n z ) is the component of the normal vector, and d is the distance from the origin to the plane. Calculate the current hole center coordinates (o x ,o y ,o z ) and the normal vector (n x ,n y ,n z ).

[0065] Step S7: according to the D-H parameter method, the homogeneous transformation matrix between the adjacent links of the robot arm 2-7 is obtained, and the homogeneous transformation matrix between the adjacent links of the robot arm 2-7 is multiplied to obtain the pose equation of the end effector relative to the base. According to the obtained hole site pose information, the inverse kinematics of the robot arm 2-7 is solved, and the rotation angles θ1, θ2, θ3, θ4 of the joints of the robot arm 2-7 are calculated. Adjust the pose of the end effector of the robot arm 2-7 to be perpendicular to the hole site.

[0066] The homogeneous coordinate transformation represents the coordinate transformation relationship between adjacent links of the robot arm:

[0067]

[0068] where i represents each joint of the robot arm; θ i represents the angle between the x i axis and the x i-1 axis in the z i axis direction; d i represents the distance between the x i axis and the x i-1 axis in the z i axis direction; a i-1 represents the distance between the x i-1The axis direction, z i-1 The axis and z i The distance between the axes; α i-1 Indicates that the x i-1 The axis direction, z i-1 The axis and z i The angle between the axes.

[0069] According to the D-H parameter rule, the final matrix is calculated:

[0070]

[0071] The joint rotation angles are calculated by the inverse kinematics of the robot arm as follows:

[0072] θ1=arctan2(p y ,p x )

[0073]

[0074] θ3=-arccos(-a z )-θ2

[0075]

[0076] where [p x p y p z ] T represents the spatial position of the robot arm end in the base coordinate system, represents the robot arm end posture.

[0077] Step S8: The compliant installation of the water stop needle 3-4 insertion hole is completed in combination with the admittance control, the robot arm 2-7 end effector pose is constantly fine-tuned according to the six-dimensional force sensor 3-2, and the blocking phenomenon is eliminated; the principle of the admittance control is that the control system converts the external force into the deviation of the relative expected displacement, the robot arm 2-7 end is subjected to the z-axis external force F ext , and the initial expected trajectory of the robot arm 2-7 end is x0. When the robot arm 2-7 is subjected to the external force, it will be offset on the original trajectory to comply with the external force, and a new expected position x d is generated by the admittance controller, and the control system formula is:

[0078]

[0079] where M d , D d , K d are control system parameters. The compliant assembly control of the water stop needle 3-4 insertion hole is realized by constantly correcting the position deviation, and the blocking phenomenon during assembly is eliminated.

[0080] Step S9: When the six-dimensional force sensor 3-2 at the end of the mechanical arm 2-7 measures that the assembly direction z-axis force is greater than a specified threshold value, and the hole depth and the insertion depth of the water stop needle 3-4 are less than a specified threshold value, the water stop needle 3-4 is successfully installed, and the step ends; otherwise, continue with steps S7-S8.

[0081] The method is based on the CycleGAN to restore underwater images and combines the deep learning method Yolov5 to complete hole position detection; a parallax map is obtained by combining a binocular stereo matching method to complete three-dimensional point cloud reconstruction; according to hole position area information, the RANSAC method is used to complete accurate segmentation of hole position point cloud and calculation of hole position pose information; the method improves the detection speed and the accuracy of the pose information, can quickly complete the tunnel concrete lining crack plugging work, improves the automation level and safety of the water stop needle remote installation operation of the tunnel concrete lining crack, and significantly improves the operation efficiency.

[0082] It should be noted that the above content only illustrates the technical idea of the present application and cannot limit the protection scope of the present application. For ordinary skilled persons in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which fall within the protection scope of the claims of the present application.

Claims

1. A water stop needle remote installation system for underwater crack repair of a tunnel, characterized in that: the underwater robot comprises a master end, a communication channel and a slave end, the master end is connected to the slave end underwater robot through the communication channel and communicates with the slave end underwater robot, and controls the slave end to work. The slave end underwater robot comprises at least a monocular camera and left and right binocular cameras, the monocular camera is located at the center of the top of the underwater robot, and the binocular cameras are located at the end of the mechanical arm of the underwater robot. The master end comprises at least a force feedback hand controller and a remote control terminal, the remote control terminal restores underwater images based on a cycle generation adversarial network CycleGAN according to left and right eye images of the binocular camera, inputs original underwater images into a network structure to obtain clear underwater images, manually labels underwater dataset pictures based on a deep learning method Yolov5 to train the network, inputs the restored left eye pictures into the trained network model to obtain hole position anchor frames, completes hole position detection, obtains a disparity map by combining a binocular stereo matching method SGBM, compensates for holes in the disparity map, completes three-dimensional point cloud reconstruction, removes outliers by using a Gaussian filter, fits hole position point cloud information by using a RANSAC algorithm, calculates a hole position plane, obtains hole center point coordinates and a normal vector, adjusts the end of the mechanical arm to be perpendicular to the hole position according to hole position pose information, and installs the water stop needle.

2. The waterstop needle remote installation system for underwater tunnel crack repair of claim 1, wherein: The end of the underwater robot mechanical arm further comprises a six-dimensional force sensor, the six-dimensional force sensor is used for measuring contact force between the water stop needle and the hole and adjusting the posture of the end of the mechanical arm, and the mechanical arm is provided with a waterproof sealing layer.

3. A water stop needle remote installation system for underwater crack repair of a tunnel according to claim 1 or 2, characterized in that: The monocular camera and the binocular cameras are movably connected to the underwater robot and can move up and down and rotate, and the monocular camera and the binocular cameras are provided with waterproof protective shells outside.

4. A method for remote installation of a water stop needle for underwater crack repair of a tunnel using the system as claimed in claim 1, characterized in that: The method comprises the following steps: S1: The underwater environment video information is fed back to the remote control terminal and displayed through the monocular camera of the slave end, an operator observes underwater cracks and hole positions, and determines a preliminary hole insertion position of the water stop needle; S2: The operator remotely controls the end of the underwater mechanical arm to approach the preliminarily selected hole position through the force feedback hand controller; S3: The left and right eye images of the binocular camera arranged at the end of the mechanical arm are used to restore underwater images based on a cycle generation adversarial network CycleGAN, original underwater images are input into a network structure to obtain clear underwater images, underwater dataset pictures are collected in advance, the underwater dataset pictures after clarity restoration are manually labeled, hole position regions are labeled on the pictures by using a label frame, a dataset is trained based on a deep learning method Yolov5, a trained model is obtained, when the left eye picture after CycleGAN network restoration is obtained, the Yolov5 network structure is input into the trained network structure, hole position anchor frames are obtained, and hole position detection is completed; S4: A disparity map is obtained by combining a binocular stereo matching method SGBM, holes in the disparity map are compensated, three-dimensional point cloud reconstruction is completed, outliers are removed by using a Gaussian filter, hole position point cloud information is fitted by using a RANSAC algorithm, a hole position plane is calculated, hole center point coordinates and a normal vector are obtained, the end of the mechanical arm is adjusted to be perpendicular to the hole position according to hole position pose information, and the water stop needle is installed. S5: transmitting the three-dimensional point cloud obtained in step S4 to the remote control terminal through the communication channel and displaying on the visualization interface, Gaussian filtering is performed on the global point cloud to remove outliers, and at the same time, according to the hole site information obtained in step S3, the hole site point cloud is segmented out; S6: fitting the hole site point cloud information using the RANSAC algorithm, calculating the current hole center coordinate and normal vector, the calculation method is, according to the point cloud information, the plane equation H of the hole is solved: H: n x x + n y y + n z z + d = 0 wherein (n x , n y , n z ) are components of the normal vector, and d is the distance from the origin to the plane; according to the plane equation H and the point cloud result, the current hole center coordinates (o x , o y , o z ) and the normal vector (n x , n y , n z ) are calculated; S7: according to the D-H parameter method, the homogeneous transformation matrix between adjacent links of the mechanical arm is solved, the homogeneous transformation matrix between each link of the mechanical arm is multiplied to obtain the pose equation of the end effector relative to the base; according to the hole pose information obtained in step S6, inverse kinematics is solved for the mechanical arm pose information, the rotation angles θ1, θ2, θ3, θ4 of each joint of the mechanical arm are calculated, the pose of the end effector of the mechanical arm is adjusted to be perpendicular to the hole site, and the installation of the water stop needle head is prepared; S8: combining the admittance control to complete the compliant installation of the water stop needle head, according to the six-dimensional force sensor, the pose of the end effector of the mechanical arm is constantly fine-tuned to eliminate the blocking phenomenon; S9: when the six-dimensional force sensor at the end of the mechanical arm measures that the assembly direction z-axis force is greater than a specified threshold value, and the hole depth and the insertion depth of the water stop needle head are less than a specified threshold value, the water stop needle head is successfully installed, and the step is ended; otherwise, steps S7-S8 are continued.

5. The method for remote installation of a waterstop needle for underwater tunnel crack repair of claim 4, wherein: In step S2, the force feedback hand controller has 3 degrees of freedom, 3 joint degrees of freedom and 3 degrees of freedom force feedback output, adopts a position mapping mechanism to map the spatial position at the end of the force feedback hand controller to the end of the mechanical arm, and the mapping relationship is: Wherein, [x p , y p , z p ] is the force feedback hand controller end position, [x, y, z] is the underwater manipulator end position, a x , a y , a z are the amplification multiples along the x-axis, y-axis and z-axis directions respectively.

6. The method for remote installation of a waterstop needle for underwater tunnel crack repair of claim 5, wherein: The step S5 is to remove outliers by Gaussian filtering on the global point cloud, and k neighborhood points around each point cloud data are searched based on KD-Tree, and the average Euclidean distance from the sampling point to the k neighborhood points is calculated, and p i = {x i , x i , x i | i = 1, 2, 3,..., n} is the original point cloud data, and the data set after KD-Tree search is p j = {x j , x j , x j | j = 1, 2, 3,..., k}, and the following formula is obtained: where d i is the average distance of the point p j to its k neighborhood points, and is the mean value of d i with the standard deviation σ, given its mean and variance, outliers of the point cloud are removed.

7. A method for remote installation of a waterstop needle for underwater crack repair of a tunnel according to claim 5 or 6, characterized in that: In step S7, the calculation method of the rotation angles θ1, θ2, θ3, θ4 of each joint of the mechanical arm is: By homogeneous coordinate transformation Represent the coordinate transformation relationship between adjacent links of the robot arm: where i represents each joint of the robot arm; θ i represents the angle between the z i axis and the x i-1 axis as viewed in the z i axis; d i represents the distance between the z i axis and the x i-1 axis as viewed in the z i axis; a i-1 represents the distance between the x i-1 axis and the z i-1 axis as viewed in the x i axis; α i-1 represents the angle between the x i-1 axis and the z i-1 axis as viewed in the x i axis; According to the D-H parameter rule, the final matrix is calculated: Through the inverse kinematics of the mechanical arm, the rotation angles of each joint are respectively: θ1 = arctan2(p y , p x ) θ3 = -arccos(-a z )-θ2 wherein [p x p y p z ] T denotes the spatial position of the robot arm end in the base coordinate system, denotes the robot arm end pose.

Images