A tunnel boring machine positioning device and method based on active vision

Abstract

The present invention discloses an active vision-based positioning device and method for a tunnel boring machine. The device comprises a tunnel boring machine, an active vision module, an artificial feature object, a total station, and a positioning control system. The active vision module is mounted on the tunnel boring machine, and the artificial feature object is positioned between the active vision module and the total station. The positioning control system is used to control the active vision module to acquire images of the artificial feature object at different positions and viewing angles, and directly estimate the pose parameters of an industrial camera in the active vision module relative to the artificial feature object based on the images. Combining the pose state of the artificial feature object in the tunnel and the pose state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom pose of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation. The total station is used to reposition the artificial feature object after it is moved, thereby achieving autonomous and continuous positioning of the tunnel boring machine. The present invention has the advantages of high robustness, high precision, good real-time performance, low cost, and simple system construction.

Description

Technical Field

[0001] The present invention relates to a tunnel boring machine positioning technology, and in particular to a tunnel boring machine positioning device and method based on active vision. Background Art

[0002] Coal has long occupied a dominant position in my country's energy consumption structure and is a vital pillar of the national economy. 90% of my country's total coal production comes from underground mining, which requires extensive tunneling. Currently, my country's coal mine tunneling exceeds 12,000 kilometers annually. Tunneling is a high-risk industry with frequent accidents. The realization of unmanned or reduced-manpower coal mine tunneling is an inevitable trend in the development of the coal mining industry. Autonomous positioning of tunnel boring machines (TBMs) is a primary requirement for intelligent tunneling and a key issue limiting the level of TBM intelligence. TBMs operate in narrow, enclosed underground spaces, lacking GPS signals. The tunneling environment is complex, characterized by weak and unstructured textures, complex electromagnetic environments, variable lighting conditions, high dust and water vapor concentrations, numerous electromechanical equipment, and complex scheduling. Furthermore, TBMs spend extended periods of stationary operation accompanied by intense vibration. Consequently, accurate and real-time positioning of TBMs within tunnels is challenging. Currently, the main positioning methods include total stations, wireless sensor networks (WSNs) or ultra-wideband (UWB)-based radio transceiver technologies, inertial navigation, lidar SLAM (LiDAR), machine vision, and multi-sensor combinations. However, due to the extreme environment of underground coal mines and the complex working conditions of tunneling, these machine body posture detection methods have certain problems or difficulties in actual application in coal mine fully mechanized excavation working faces. Therefore, it is urgent to research positioning methods and strategies suitable for roadheaders. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a new type of tunnel boring machine positioning device based on active vision.

[0004] Another object of the present invention is to provide a robust, accurate, low-cost, and highly real-time method for positioning a tunnel boring machine based on active vision, which effectively solves the problem of autonomous positioning of the tunnel boring machine.

[0005] Technical solution: The present invention provides a tunnel boring machine positioning device based on active vision, comprising: a tunnel boring machine, an active vision module, an artificial feature object, a total station, and a positioning control system; wherein:

[0006] The active vision module is installed on the tunnel boring machine and includes a linear motor, a first rotary motor, a second rotary motor, and an industrial camera. The linear motor is used to achieve lateral movement of the industrial camera, while the first rotary motor and the second rotary motor are used to achieve rotation and pitch movement of the industrial camera respectively.

[0007] The artificial feature object is set between the active vision module and the total station at a preset distance from the active vision module. It includes a mounting plate, a total station prism, and an infrared laser array. The mounting plate is installed on the roadway roof. The side facing the active vision module and the side facing the total station are respectively equipped with an infrared laser array and a total station prism. The infrared laser array provides rich structured features for the industrial camera to achieve visual positioning of the roadheader. The total station prism is used in conjunction with the total station.

[0008] The total station is placed at the rear of the tunnel to provide the tunnel's absolute coordinate system. The total station's prism is used to measure the position of artificial feature objects in the tunnel. The total station is also used to reposition the artificial feature objects after they move forward, enabling long-distance continuous positioning of the tunnel boring machine.

[0009] The positioning control system includes: a camera control system and a tunnel boring machine posture estimation system. The camera control system is based on the reinforcement learning method to establish a camera action decision reinforcement learning model, and controls the industrial camera to obtain images of artificial feature objects at different positions and perspectives by driving the linear motor and the first and second rotary motors; the posture state of the industrial camera relative to the tunnel boring machine is determined according to the running trajectory of the three motors; the tunnel boring machine posture estimation system is based on the deep learning method to construct a camera posture regression deep learning model, and directly estimates the posture parameters of the industrial camera relative to the artificial feature object based on the image obtained by the industrial camera. Combined with the posture state of the artificial feature object in the tunnel and the posture state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom posture of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

[0010] Preferably, the three motors are configured with encoders respectively to obtain the lateral displacement, rotation angle and pitch angle of the industrial camera, and then determine the position state of the industrial camera relative to the tunnel boring machine.

[0011] Preferably, the industrial camera is an autofocus industrial camera with a smaller field of view and a larger aperture.

[0012] Preferably, as the tunnel boring machine advances forward, the relative distance between the artificial feature object and the active vision module is maintained within a preset range.

[0013] In another embodiment of the present invention, a method for positioning a tunnel boring machine based on active vision includes the following steps:

[0014] S1. Deep learning model training phase: In an underground tunnel environment simulated in an above-ground laboratory, an active vision module is deployed on a mobile machine, and artificial feature objects are placed behind the machine. The pose parameters of the active vision module relative to the artificial feature objects are obtained online to obtain rich camera image-camera pose data pairs for training the camera action decision reinforcement learning model and the camera pose regression deep learning model.

[0015] S2. Complete the initial arrangement of the active vision module, artificial feature objects, and total station in the tunnel to be excavated, and measure the position parameters of the artificial feature objects in the tunnel;

[0016] S3: The tunnel boring machine begins excavation. The trained camera action decision reinforcement learning model is used to control the active vision module in real time, allowing the industrial camera to capture images of the artificial feature objects at a preferred position and viewing angle. The pose parameters of the industrial camera on the tunnel boring machine are calculated in real time based on the values ​​of the motor encoder. The trained camera pose regression deep learning model is used to estimate the pose parameters of the industrial camera relative to the artificial feature objects, thereby achieving pose estimation of the tunnel boring machine in the absolute coordinate system of the tunnel.

[0017] S4. As the roadheader advances, the artificial feature objects in the images captured by the industrial camera will become smaller and smaller, and the features will become less and less obvious. That is, when the artificial feature objects exceed the effective range of the industrial camera's visual positioning, the accuracy of the pose estimation will be affected. The artificial feature objects need to be moved forward and their pose parameters in the tunnel need to be remeasured.

[0018] S5. Loop steps S3-S4 to achieve online estimation of the machine body posture during long-distance operation of the tunnel boring machine, providing conditions for intelligent tunneling.

[0019] Furthermore, when training the model in an experimental environment, the camera pose regression deep learning model is first trained offline to estimate the pose parameters of the industrial camera relative to the artificial feature objects; then the camera action decision reinforcement learning model is trained online to enable the camera to observe the artificial feature objects from a better perspective; finally, the two models are jointly trained and fine-tuned to improve model performance.

[0020] Furthermore, the method is also applicable to rock drilling rigs and continuous coal mining machines.

[0021] In yet another embodiment of the present invention, a tunnel boring machine positioning system based on active vision includes:

[0022] The data acquisition module is used to collect images of the artificial feature object and data from three motor encoders to measure the position and state of the artificial feature object in the lane;

[0023] The camera action decision module is used to control the industrial camera to acquire images of artificial feature objects at different positions and viewing angles, and to determine the position and posture of the industrial camera relative to the tunnel boring machine based on the encoder data;

[0024] The tunnel boring machine pose estimation module is used to directly estimate the pose parameters of the industrial camera relative to the artificial feature object based on the image of the artificial feature object acquired by the industrial camera. Combining the pose state of the artificial feature object in the tunnel and the pose state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom pose of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

[0025] In yet another embodiment of the present invention, an electronic device includes a memory and a processor, wherein:

[0026] a memory for storing computer programs capable of running on the processor;

[0027] The processor is configured to execute the steps of the above-mentioned method for positioning a tunnel boring machine based on active vision when running the computer program.

[0028] In another embodiment of the present invention, a storage medium stores a computer program, and when the computer program is executed by at least one processor, the steps of the above-mentioned active vision-based positioning method for a tunnel boring machine are implemented.

[0029] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are: it has the advantages of convenient deployment, low cost, robustness and precision, and good real-time performance. Through the active vision module, based on reinforcement learning technology, the viewing angle and position of the industrial camera are controlled in real time to avoid visual occlusion, reduce image distortion, improve positioning accuracy and stability, and reduce the training difficulty of the camera pose regression deep learning model (after using active vision, the pose data range of the industrial camera relative to the artificial feature object can be reduced during the model training data acquisition process of the ground experiment). Estimating the pose parameters of the industrial camera relative to the artificial feature object based on the end-to-end deep learning method is conducive to improving the robustness of positioning. This method is applicable to straight tunnels and also to curved tunnels with smaller curvature. This method can be used for tunneling equipment such as cantilever tunneling machines, rock drilling rigs, and continuous coal mining machines. By solving the problem of real-time estimation of the posture of the tunneling machine body, conditions are provided for intelligent operation of the tunneling working face. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the structure of the device of the present invention;

[0031] Figure 2 is a schematic diagram of the active vision module in the present invention;

[0032] Figure 3 Schematic diagram of an artificial feature object in the present invention, wherein (a) is a schematic diagram of the reverse side and (b) is a schematic diagram of the front side;

[0033] Figure 4 It is a flow chart of the method of the present invention;

[0034] In the figure: 1. Active vision module; 2. Artificial feature object; 3. Total station; 1-1. Linear motor; 1-2. First rotary motor; 1-3. Second rotary motor; 1-4. Industrial camera; 2-1. Mounting bracket; 2-2. Mounting plate; 2-3. Total station prism; 2-4. Infrared laser array. DETAILED DESCRIPTION

[0035] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown in the figure, an active vision-based positioning device for a tunnel boring machine of the present invention comprises: a tunnel boring machine, an active vision module 1, an artificial feature object 2, a total station 3, and a positioning control system (an external device not shown). The active vision module 1 is mounted on the tunnel boring machine, the artificial feature object 2 is mounted on the tunnel roof a distance behind the tunnel boring machine, and the total station 3 is arranged at the rear of the tunnel to provide an absolute coordinate system for the tunnel. The artificial feature object 2 is located between the active vision module 1 and the total station 3. As the tunnel boring machine advances, the relative distance between the artificial feature object and the active vision module remains within a preset range (approximately 10 to 25 meters).

[0037] like Figure 2 As shown, the active vision module 1 includes a linear motor 1-1, a first rotary motor 1-2, a second rotary motor 1-3 and an industrial camera 1-4. The linear motor is used to realize the lateral movement of the industrial camera, and the two rotary motors are used to realize the rotation and pitch movement of the industrial camera respectively. Through the drive of the motors, the industrial camera 1-4 can be controlled to obtain images of the artificial feature object 2 at different positions and viewing angles, and then the posture parameters of the industrial camera 1-4 relative to the artificial feature object 2 are estimated. The three motors are all equipped with encoders, and the lateral displacement, rotation angle and pitch angle of the industrial camera are obtained through the encoders to determine the posture state of the industrial camera 1-4 relative to the tunnel boring machine.

[0038] In the embodiment of the present invention, in order to adapt to the working environment of the tunnel and increase the effective positioning range when the artificial feature object 2 is fixed, the industrial cameras 1-4 use industrial cameras with a smaller field of view angle (about 24° to 34°) and a larger aperture (about f / 4); at the same time, industrial cameras with automatic focus are selected to ensure that clear images are obtained at any object distance.

[0039] like Figure 3As shown, the artificial feature object 2 includes a mounting frame 2-1, a mounting plate 2-2, a total station prism 2-3, and an infrared laser array 2-4; the infrared laser array 2-4 is installed on the front side of the mounting plate 2-2 (i.e., the side facing the active vision module), providing rich structured features for the industrial camera 1-4 to realize the visual positioning of the tunnel boring machine. The infrared laser has good penetrability in the tunnel environment with a high dust concentration; the total station prism 2-3 is installed on the back side of the mounting plate 2-2 (i.e., the side facing the total station), and is used in conjunction with the total station 3. The total station can measure the posture state of the artificial feature object 2 in the tunnel through the total station prism, and reposition the artificial feature object 2 after moving forward, so as to realize long-distance continuous positioning of the tunnel boring machine.

[0040] The positioning control system includes: a camera control system and a tunnel boring machine posture estimation system. The camera control system is based on the reinforcement learning method to establish a camera action decision reinforcement learning model. By driving the linear motor and the first and second rotary motors, the position and angle of the industrial camera are controlled to observe the artificial feature objects with a better perspective to avoid visual occlusion, reduce image distortion, and improve positioning accuracy; the tunnel boring machine posture estimation system is based on the deep learning method to construct a camera posture regression deep learning model. The image acquired by the camera directly estimates the posture parameters of the industrial cameras 1-4 relative to the artificial feature object 2. Combined with the posture state of the artificial feature object 2 in the tunnel and the posture state of the industrial cameras 1-4 relative to the tunnel boring machine, the absolute six-degree-of-freedom posture of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

[0041] Given the known poses of the artificial feature object 2 in the tunnel and the poses of the industrial cameras 1-4 relative to the tunnel boring machine, the tunnel boring machine localization problem can be transformed into estimating the pose parameters of the industrial cameras 1-4 relative to the artificial feature object 2. Therefore, an underground tunnel environment can be simulated in an above-ground laboratory. An active vision module 1 is deployed on a mobile machine, and the artificial feature object 2 is positioned behind it. Other methods (such as a motion capture system) are used to obtain the pose parameters of the active vision module relative to the artificial feature object. This yields a wealth of camera image-camera pose data pairs for training the camera action decision reinforcement learning model and the camera pose regression deep learning model. The trained models can be directly deployed during actual tunnel boring machine operation. During model training in the experimental environment, the camera pose regression deep learning model is first trained offline to estimate the pose parameters of the industrial cameras relative to the artificial feature object. Then, the camera action decision reinforcement learning model is trained online to ensure that the camera observes the artificial feature object from a better perspective, avoiding visual occlusion, reducing image distortion, and improving localization accuracy. Finally, the two models are jointly trained and fine-tuned to improve model performance.

[0042] like Figure 4As shown, a tunnel boring machine positioning method based on active vision of the present invention includes the following steps:

[0043] S1, deep learning model training phase: In an underground tunnel environment simulated in an above-ground laboratory, an active vision module 1 is configured on a mobile body, and an artificial feature object 2 is placed behind the body. The pose parameters of the active vision module relative to the artificial feature object are obtained online to obtain rich camera image-camera pose data pairs for training the camera action decision reinforcement learning model and the camera pose regression deep learning model;

[0044] The present invention trains a camera action decision reinforcement learning model based on the SAC (Soft Actor-Critic) algorithm; constructs a camera pose regression deep learning model based on the Transformer model, and uses the acquired data for end-to-end training.

[0045] During model training in an experimental environment, a camera pose regression deep learning model was first trained offline using a large amount of acquired camera image and camera pose data, achieving end-to-end estimation of the pose parameters of the industrial camera relative to artificial feature objects. Then, based on the SAC reinforcement learning method, the camera action decision reinforcement learning model was trained online, using the completeness of the artificial feature objects in the image and their proximity to the image center as reward signals, enabling the camera to observe the artificial feature objects from a better perspective. Finally, the two models were jointly trained and fine-tuned to improve model performance.

[0046] S2, the preparation stage, completes the initial arrangement of the active vision module, artificial feature objects and total station in the tunnel to be excavated, and measures the position parameters of the artificial feature objects in the tunnel;

[0047] S3. The tunnel boring machine starts excavation and uses the trained camera action decision reinforcement learning model to control the active vision module 1 in real time. According to the current image state (occlusion status, position of the artificial feature object in the image) acquired by the camera, the industrial cameras 1-4 can actively select a better position and viewing angle to acquire the image of the artificial feature object 2, and calculate the pose parameters of the industrial cameras 1-4 on the tunnel boring machine in real time based on the numerical value of the motor encoder. Using the image of the artificial feature object currently acquired by the industrial cameras, the pose parameters of the industrial cameras 1-4 relative to the artificial feature object 2 are estimated based on the trained camera pose regression deep learning model, thereby realizing the pose estimation of the tunnel boring machine in the absolute coordinate system of the tunnel;

[0048] S4. As the roadheader advances, the artificial feature object in the image captured by the industrial camera will become smaller and smaller, and the features will become less and less obvious. That is, when the artificial feature object exceeds the effective range of the industrial camera's visual positioning (about 10 to 25 meters), the accuracy of the pose estimation is affected. It is necessary to move the artificial feature object 2 forward and remeasure the pose parameters of the artificial feature object 2 in the tunnel.

[0049] By properly configuring the parameters of the industrial camera and artificial feature objects, a single station move of the artificial feature object can complete approximately 15 meters of tunnel excavation. Because tunnel excavation speed is very slow, especially in hard rock tunnels, moving the station every certain distance has a minimal impact on tunneling efficiency.

[0050] S5. Loop steps S3-S4 to achieve online estimation of the machine body posture during long-distance operation of the tunnel boring machine, providing conditions for intelligent tunneling.

[0051] The above method can be used for tunnel excavation equipment such as cantilever roadheaders, rock drilling rigs, and continuous coal mining machines.

[0052] Based on the same inventive concept, the present invention provides a tunnel boring machine positioning system based on active vision, comprising:

[0053] The data acquisition module is used to collect images of the artificial feature object and data from three motor encoders to measure the position and posture of the artificial feature object in the lane;

[0054] The camera action decision module is used to control the industrial camera to acquire images of artificial feature objects at different positions and viewing angles, and to determine the position and posture of the industrial camera relative to the tunnel boring machine based on the encoder data;

[0055] The tunnel boring machine pose estimation module is used to directly estimate the pose parameters of the industrial camera relative to the artificial feature object based on the image of the artificial feature object acquired by the industrial camera. Combining the pose state of the artificial feature object in the tunnel and the pose state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom pose of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

[0056] Based on the same inventive concept, an electronic device of the present invention includes a memory and a processor, wherein:

[0057] a memory for storing computer programs capable of running on the processor;

[0058] The processor is configured to execute the steps of the above-mentioned method for positioning a tunnel boring machine based on active vision when running the computer program.

[0059] Based on the same inventive concept, the present invention provides a storage medium having a computer program stored thereon, which, when executed by at least one processor, implements the steps of the above-mentioned active vision-based tunnel boring machine positioning method.

Claims

1. A tunnel boring machine positioning device based on active vision, characterized in that: include: A tunnel boring machine, an active vision module (1), an artificial feature object (2), a total station (3), and a positioning control system; wherein: An active vision module (1) is installed on a tunnel boring machine and includes a linear motor (1-1), a first rotary motor (1-2), a second rotary motor (1-3), and an industrial camera (1-4). The linear motor (1-1) is used to realize the lateral movement of the industrial camera (1-4), and the first rotary motor (1-2) and the second rotary motor (1-3) are used to realize the rotation and pitch movement of the industrial camera (1-4), respectively. An artificial feature object (2) is set between the active vision module (1) and the total station (3) at a preset distance from the active vision module (1), and includes a mounting plate (2-2), a total station prism (2-3) and an infrared laser array (2-4). The mounting plate (2-2) is mounted on the top plate of the tunnel, and an infrared laser array (2-4) and a total station prism (2-3) are mounted on the side facing the active vision module (1) and the side facing the total station (3), respectively. The infrared laser array provides rich structured features for the industrial camera to realize visual positioning of the tunnel boring machine, and the total station prism is used in conjunction with the total station. The total station (3) is arranged at the rear of the tunnel and is used to provide an absolute coordinate system for the tunnel. The position and posture of the artificial feature object in the tunnel are measured by the cooperation of the total station prism, and the artificial feature object is used to reposition after moving forward, so as to realize long-distance continuous positioning of the tunnel boring machine. The positioning control system includes: a camera control system and a tunnel boring machine posture estimation system. The camera control system is based on the reinforcement learning method to establish a camera action decision reinforcement learning model, and controls the industrial camera to obtain images of artificial feature objects at different positions and perspectives by driving the linear motor and the first and second rotary motors; the posture state of the industrial camera relative to the tunnel boring machine is determined according to the running trajectory of the three motors; the tunnel boring machine posture estimation system is based on the deep learning method to construct a camera posture regression deep learning model, and directly estimates the posture parameters of the industrial camera relative to the artificial feature object based on the image obtained by the industrial camera. Combined with the posture state of the artificial feature object in the tunnel and the posture state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom posture of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

2. The active vision-based positioning device for a tunnel boring machine according to claim 1, characterized in that: The three motors are equipped with encoders to obtain the lateral displacement, rotation angle and pitch angle of the industrial camera, and then determine the position state of the industrial camera relative to the tunnel boring machine.

3. The active vision-based positioning device for a tunnel boring machine according to claim 1, characterized in that: The industrial camera should be an autofocus industrial camera with a field of view and aperture within the set range.

4. The active vision-based positioning device for a tunnel boring machine according to claim 1, characterized in that: As the tunnel boring machine advances, the relative distance between the artificial feature object and the active vision module remains within a preset range.

5. A tunnel boring machine positioning method based on active vision, characterized in that: The following steps are involved: S1, deep learning model training phase, simulates an underground tunnel environment in an above-ground laboratory, configures an active vision module (1) on a mobile body, configures an artificial feature object (2) behind the body, and obtains the pose parameters of the active vision module relative to the artificial feature object online to obtain rich camera image-camera pose data pairs for training camera action decision reinforcement learning models and camera pose regression deep learning models; S2. Complete the initial arrangement of the active vision module, artificial feature objects, and total station in the tunnel to be excavated, and measure the position parameters of the artificial feature objects in the tunnel; S3, the tunnel boring machine starts excavation work, and uses the trained camera action decision reinforcement learning model to control the active vision module (1) in real time, so that the industrial camera (1-4) can obtain the image of the artificial feature object (2) at a better position and viewing angle, and calculates the posture parameters of the industrial camera (1-4) on the tunnel boring machine in real time based on the numerical value of the motor encoder, and uses the trained camera posture regression deep learning model to estimate the posture parameters of the industrial camera (1-4) relative to the artificial feature object (2), thereby realizing the posture estimation of the tunnel boring machine in the absolute coordinate system of the tunnel; S4. As the tunnel boring machine advances forward, the artificial feature object (2) exceeds the effective range of the industrial camera visual positioning, and the artificial feature object (2) needs to be moved forward and the posture parameters of the artificial feature object (2) in the tunnel need to be remeasured; S5. Loop steps S3-S4 to achieve online estimation of the machine body posture during long-distance operation of the tunnel boring machine, providing conditions for intelligent tunneling.

6. The method for positioning a tunnel boring machine based on active vision according to claim 5, characterized in that: When training the model in an experimental environment, the camera pose regression deep learning model is first trained offline to estimate the pose parameters of the industrial camera relative to the artificial feature objects; then the camera action decision reinforcement learning model is trained online to enable the camera to observe the artificial feature objects from a better perspective; finally, the two models are jointly trained and fine-tuned to improve model performance.

7. The method for positioning a tunnel boring machine based on active vision according to claim 6, characterized in that: This method is also applicable to rock drilling rigs and continuous coal mining machines.

8. A system used in the active vision-based positioning method for a roadheader according to any one of claims 5 to 7, characterized in that: include: The data acquisition module is used to collect images of the artificial feature object and data from three motor encoders to measure the position and state of the artificial feature object in the lane; The camera action decision module is used to control the industrial camera to acquire images of artificial feature objects at different positions and viewing angles, and to determine the position and posture of the industrial camera relative to the tunnel boring machine based on the encoder data; The tunnel boring machine pose estimation module is used to directly estimate the pose parameters of the industrial camera relative to the artificial feature object based on the image of the artificial feature object acquired by the industrial camera. Combining the pose state of the artificial feature object in the tunnel and the pose state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom pose of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: a memory for storing computer programs capable of running on the processor; A processor is configured to execute the steps of a tunnel boring machine positioning method based on active vision as described in any one of claims 5 to 7 when running the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by at least one processor, implements the steps of the active vision-based tunnel boring machine positioning method as described in any one of claims 5 to 7.

Images