Tunnel convergence deformation detection method and system based on 3D camera
The tunnel convergence deformation detection method based on 3D cameras solves the problems of low efficiency, limited accuracy and low point cloud generation density of traditional lidar detection, and realizes efficient and automated tunnel convergence deformation detection, meeting the precise requirements of subway safe operation.
Patent Information
- Application Number
- CN202511683429.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-02-06
AI Technical Summary
Traditional lidar detection of deformation in subway shield tunnels requires manual intervention, has low detection efficiency, low point cloud generation density, and limited accuracy, and cannot meet the needs of rapid and accurate detection for long-distance lines.
A tunnel convergence deformation detection method based on 3D cameras is adopted. Multiple 3D cameras collect point cloud data in real time. Combined with photoelectric encoders and vehicle posture compensation devices, the system corrects deviations and optimizes the center positioning in real time, generating denser point cloud data for filtering and feature recognition, thereby achieving automated detection.
It achieves automated detection without human intervention, improving detection efficiency and accuracy, generating denser point cloud data, meeting the needs of rapid detection of long-distance lines, and achieving an accuracy close to the ±5mm tunnel convergence detection standard, providing more reliable safety assurance.
Smart Images
Figure CN121482004A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of structural deformation detection, specifically to a method and system for detecting tunnel convergence deformation based on a 3D camera. Background Technology
[0002] Subway shield tunnels are underground tunnel structures excavated using tunnel boring machines (TBMs). As these tunnels age, deformation (such as settlement, heave, and cracking) can lead to track deviations, and in severe cases, train derailments and tunnel collapses. Therefore, in daily operation and maintenance, it is crucial to closely monitor and regularly review the tunnel's convergence status, accurately identify abnormal sections, and implement proactive prevention measures. This is key to ensuring the safe operation of the subway. Currently, the mainstream technology for detecting deformation in subway shield tunnels typically uses traditional multi-line rotating lasers to perform three-dimensional scanning of the tunnel, thereby detecting tunnel convergence deformation. However, traditional lidar detection has the following drawbacks: Manual intervention is required, and the debugging process is cumbersome: during testing, the equipment must be placed in the track area, and several people must complete the equipment debugging and preparation before the tunnel cross-section can be tested. In addition, the equipment must be calibrated before each test.
[0003] Low inspection efficiency: The normal inspection speed of the rotating laser tunnel 3D scanning inspection equipment is about 5 km / h, and only about 5 km can be inspected each night. It cannot meet the requirements for rapid inspection of long-distance lines.
[0004] Low point cloud generation density: Taking a 128-line lidar as an example, it generates approximately 1.152 million points per second, which is insufficient for the structure of a subway shield tunnel. Therefore, with the reduction in raw point cloud data, multiple detections or a reduction in detection speed are necessary to ensure detection accuracy.
[0005] Limited detection accuracy: The detection accuracy of lidar is usually 5-30mm, while the detection standard for tunnel convergence is ±5mm. The existing accuracy basically meets the minimum requirements, but further improvement is still needed to achieve reliable application in tunnel convergence detection. Summary of the Invention
[0006] In view of this, the present invention provides a tunnel convergence deformation detection method based on a 3D camera, comprising the following steps: S1: Real-time acquisition of point cloud data of the tunnel using multiple 3D cameras installed on the inspection vehicle; S2: Obtain the ring number information of the tunnel in real time, and associate the point cloud data with the ring number information; S3: Calculate the deflection angle of the detected vehicle in real time, and correct the deflection angle of the detected vehicle. S4: Repeat steps S1 to S3 to stitch together all the obtained point cloud data; S5: Filter and denoise the point cloud data obtained in S4; S6: Optimize the center positioning accuracy of the point cloud data obtained in S5; S7: Perform feature recognition on the point cloud data obtained in S6 to obtain the structural components of the tunnel.
[0007] According to a preferred embodiment, S1 includes: S11: Determine the data acquisition sequence of multiple 3D cameras on the inspection vehicle; S12: Detect vehicle movement and collect point cloud data of the tunnel sequentially based on the collected data; S13: Convert the point cloud data into a grayscale image.
[0008] According to a preferred embodiment, S2 includes: S21: Obtain the ring number information of the tunnel based on the grayscale image; S22: Associate the point cloud data and the ring number information according to the ring number information.
[0009] According to a preferred embodiment, S3 includes: S31: Preset the normal travel angle of the detection vehicle and calculate the deflection angle of the detection vehicle in real time; S32: Based on the normal travel angle and deflection angle, correct the deviation of the detection vehicle.
[0010] According to a preferred embodiment, S4 includes: S41: Pre-calibrate the travel distance of the detection vehicle, and repeat steps S1 to S3 continuously until the travel distance is reached; S42: When the travel distance is reached, a stop message is sent to the detection vehicle; S43: Stitch together all the obtained point cloud data.
[0011] According to a preferred embodiment, S5 includes: S51: Perform processing to remove abnormal light spots, dust, and noise; S52: Perform smoothing filtering.
[0012] According to a preferred embodiment, S6 includes: S61: Based on the point cloud data obtained in S5, obtain the corresponding two-dimensional array and grayscale image data; S62: Construct a basic variance matrix based on the two-dimensional array, construct a plane equation based on the variance matrix, perform circle fitting on the data using the least squares method to obtain the first profile point data, solve the plane equation based on the profile point data, and obtain the coordinates of the first circle center. S63: Based on the first profile point data, select data points at certain angular intervals and connect them to form multiple chords. Draw the perpendicular bisector of each chord according to its midpoint to obtain multiple sets of continuous straight lines. Connect two adjacent perpendicular bisectors and use their intersection as the basic data to construct a system of linear equations. Calculate the average value of the coordinates of all intersection points as the center coordinates of the circle, and calculate the average value of the coordinates of all points to the center coordinates of the circle as the radius to obtain the second center coordinates of the circle. S64: Based on the grayscale image data, obtain the second profile point data, and apply the Hough circle transform algorithm to the second profile point data to obtain the coordinates of the third circle center; S65: Check if the coordinates of the first, second, and third center points are collinear; if they are not collinear, connect the three points to form a triangle, and the center of its circumcircle is the fourth center; if the three points are collinear, take the midpoint of the line connecting the three points as the fourth center; calculate the average distance and standard deviation from the fourth center to all points, remove discrete data points based on the average radius and standard deviation, and then recalculate the average radius based on the remaining data points to obtain the final center. S66: Optimize the center positioning accuracy based on the final center.
[0013] According to a preferred embodiment, S7 includes: S71: Perform feature recognition and segmentation on the point cloud data obtained in S6; S72: Obtain the structural components of the tunnel and perform structural testing on the structural components.
[0014] According to a preferred embodiment, the structural detection includes ellipticity detection, convergence detection, and misalignment detection.
[0015] Accordingly, this invention also proposes a tunnel convergence deformation detection system based on a 3D camera to implement the above method, comprising: The hardware system includes: a 3D camera, photoelectric encoder, and vehicle posture compensation device installed on the detection vehicle, and multiple electronic tags installed at fixed distances in the tunnel; The 3D camera is used to collect point cloud data of the tunnel; The photoelectric encoder is fixedly installed on the axle of the detection vehicle and is used to calculate the mileage and speed, while triggering the 3D camera at a fixed frequency. The vehicle posture compensation device is located in the middle of the detection vehicle to dynamically compensate for the vehicle body deflection angle during the movement. Adjacent electronic tags should maintain a fixed distance; Software equipment systems, including data collection systems and data analysis systems; The data collection system is responsible for collecting point cloud data from the 3D camera, mileage and speed data from the photoelectric encoder, real-time vehicle attitude data provided by the vehicle attitude compensation device, and corresponding electronic tag data in real time. The data analysis system is used to analyze the data collected by the data collection system to achieve three-dimensional detection of the tunnel.
[0016] The tunnel convergence deformation detection method and system based on 3D cameras provided by this invention collects tunnel point cloud data in real time by setting up multiple 3D cameras on the inspection vehicle. This eliminates the need for manual on-track equipment deployment and repeated calibration, solving the problems of cumbersome manual intervention and debugging steps required by traditional lidar detection. Simultaneously, the detection can proceed with the vehicle's movement, enabling rapid long-distance line inspection as needed, thus addressing the issue of low detection efficiency. Collaborative acquisition by multiple 3D cameras generates denser raw point cloud data. Combined with the point cloud stitching step, it eliminates the need for multiple inspections or speed reductions to ensure data volume, solving the problem of low point cloud generation density. Furthermore, this method effectively compensates for the impact of driving deviations on the detection results by calculating and correcting the vehicle's deflection angle in real time, combined with optimization of the center positioning accuracy. This breaks through the 5-30mm accuracy limitation of traditional lidar, bringing it closer to the ±5mm tunnel convergence detection standard, solving the problem of limited detection accuracy. Overall, it boasts advantages such as high automation, high detection efficiency, excellent point cloud data quality, and detection accuracy that better meets standards, enabling more accurate monitoring of tunnel convergence status and providing more reliable assurance for the safe operation of subways. Attached Figure Description
[0017] Figure 1 This is a flowchart of a tunnel convergence deformation detection method based on a 3D camera according to this application; Figure 2 This is a diagram illustrating the components of a tunnel convergence deformation detection system based on a 3D camera, as described in this application. Figure 3 This is a precision verification statistics table of one embodiment of this application; Figure 4 This is a deviation statistics table of a circle center positioning accuracy optimization algorithm according to an embodiment of this application; Figure 5 This is a schematic diagram of the installation of a 3D camera on a vehicle according to an embodiment of this application. Detailed Implementation
[0018] Subway shield tunnels are underground tunnel structures excavated using tunnel boring machines (TBMs). As these tunnels age, deformation (such as settlement, heave, and cracking) can lead to track deviations, and in severe cases, train derailments and tunnel collapses. Therefore, in daily operation and maintenance, it is crucial to closely monitor and regularly review the tunnel's convergence status, accurately identify abnormal sections, and implement proactive prevention measures. This is key to ensuring the safe operation of the subway. Currently, the mainstream technology for detecting deformation in subway shield tunnels typically uses traditional multi-line rotating lasers to perform three-dimensional scanning of the tunnel, thereby detecting tunnel convergence deformation. However, traditional lidar detection has the following drawbacks: Manual intervention is required, and the debugging process is cumbersome: during testing, the equipment must be placed in the track area, and several people must complete the equipment debugging and preparation before the tunnel cross-section can be tested. In addition, the equipment must be calibrated before each test.
[0019] Low inspection efficiency: The normal inspection speed of the rotating laser tunnel 3D scanning inspection equipment is about 5km / h, and only about 5km can be inspected each night. It cannot meet the requirements for rapid inspection of long-distance lines.
[0020] Low point cloud generation density: Taking a 128-line lidar as an example, it generates approximately 1.152 million points per second, which is insufficient for the structure of a subway shield tunnel. Therefore, with the reduction in raw point cloud data, multiple detections or a reduction in detection speed are necessary to ensure detection accuracy.
[0021] Limited detection accuracy: The detection accuracy of lidar is usually 5-30mm, while the detection standard for tunnel convergence is ±5mm. The existing accuracy basically meets the minimum requirements, but further improvement is still needed to achieve reliable application in tunnel convergence detection.
[0022] In view of this, the present invention provides a method and system for detecting tunnel convergence deformation based on a 3D camera. To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been presented in the various embodiments of this application to facilitate a better understanding of the application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments.
[0023] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.
[0024] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.
[0025] The embodiments of this application will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of this application. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of this application, but are merely for illustrating the essential spirit of the technical solution of this application.
[0026] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.
[0027] Specifically, this application proposes a tunnel convergence deformation detection system based on a 3D camera, such as... Figure 2 As shown, the hardware system includes: a 3D camera, photoelectric encoder, and vehicle posture compensation device installed on the detection vehicle, and multiple electronic tags installed at fixed distances in the tunnel. The 3D camera is used to collect point cloud data of the tunnel; The photoelectric encoder is fixedly installed on the axle of the detection vehicle and is used to calculate the mileage and speed, while triggering the 3D camera at a fixed frequency. The vehicle posture compensation device is located in the middle of the detection vehicle to dynamically compensate for the vehicle body deflection angle during the movement. Adjacent electronic tags should maintain a fixed distance; Software equipment systems, including data collection systems and data analysis systems; The data collection system is responsible for collecting point cloud data from the 3D camera, mileage and speed data from the photoelectric encoder, real-time vehicle attitude data provided by the vehicle attitude compensation device, and corresponding electronic tag data in real time. The data analysis system is used to analyze the data collected by the data collection system to achieve three-dimensional detection of the tunnel.
[0028] In a specific embodiment, the execution flow is as follows: After the inspection vehicle stops at the starting station of the subway line, the hardware equipment system (3D camera, photoelectric encoder, vehicle attitude compensation device) and software equipment system (data collection system, data analysis system) are activated, and the system automatically enters self-test mode. If all hardware and software equipment systems successfully complete the self-test, the system proceeds to the next step; if the self-test fails (e.g., a 3D camera has no signal), the system immediately displays a "system malfunction" message via the data terminal and shows the location of the faulty component, allowing staff to quickly troubleshoot and avoid affecting the inspection accuracy due to equipment failure.
[0029] After passing the self-inspection, the test vehicle starts and runs normally at the required speed. The photoelectric encoder calculates the mileage and speed in real time and feeds back the train status information to the data analysis system through the data collection system. At the same time, the tunnel point cloud data collected by the 3D camera and the vehicle attitude data obtained by the vehicle attitude compensation device are transmitted to the data analysis system in real time for preliminary processing.
[0030] As the vehicle moves, the photoelectric encoder generates a TTL synchronization signal at a fixed frequency of 10Hz and sends it to the hardware system to ensure that the timestamps of the point cloud data and the vehicle attitude data are consistent, reducing subsequent data registration errors and ensuring that the subsequent vehicle attitude compensation device can correct the deviation in a timely manner. Electronic tags with a preset adjacent spacing of 100m in the tunnel are detected and identified by the vehicle, and their information is also synchronized to the data analysis system. The data collection system collects 3D camera data and vibration compensation data synchronized by the hardware in real time. If the data is successfully written, it is stored in the specified format; if the writing fails, the system immediately stops the acquisition and saves the acquired data to avoid data loss.
[0031] The data analysis system acquires raw 3D point cloud data, vehicle vibration compensation data, and real-time positioning data. It analyzes raw data from multiple 3D cameras to generate complete 3D point cloud data and converts the point cloud data into grayscale images, providing a clear image foundation for subsequent ring number recognition.
[0032] Based on the ring number identifier of the tunnel segment in the grayscale image, the ring number data of the current detection section is identified, and the extracted ring number information is temporarily stored in the data analysis system. After the corresponding 3D point cloud convergence data is calculated, the ring number and point cloud data are synchronously assigned.
[0033] The vehicle attitude compensation device calculates and detects the vehicle's deflection angle in real time. During the 3D point cloud data calculation process, the data is corrected and registered based on the deflection angle to reduce the impact of driving deviation on detection accuracy. At the same time, point cloud data from multiple 3D cameras are stitched together based on the tunnel profile features to form a preliminary tunnel profile. After the point cloud is stitched together, abnormal light spots and dust noise are first removed, then smoothing filtering is performed, and finally, the attitude correction and registration of the point cloud data are completed in combination with the vehicle deflection angle, making the point cloud data more consistent with the actual tunnel structure.
[0034] The corrected point cloud data is segmented into pipe segments, and structural features such as bolts and joints of the pipe segments are extracted. The convergence deformation data of the ring segment is calculated and associated with the previously identified ring number data. The positioning data corresponding to the ring number is extracted and dynamically associated with real-time mileage positioning information, so as to realize the binding of convergence data and mileage information. Subsequently, the convergence status of the corresponding location can be quickly queried through mileage.
[0035] The data analysis system simultaneously acquires electronic tag information, speed and mileage data from photoelectric encoders, and pre-imported basic line information databases. It integrates and analyzes these three types of data to calculate accurate mileage positioning information in real time. This information is transmitted to the data terminal for real-time display and is also associated with the identified ring number to ensure that the ring number corresponds one-to-one with the actual mileage.
[0036] Based on the ring number, a standard retrieval index is established for the registered positioning information. Users can then quickly retrieve all detection data for the corresponding segment using the ring number. At the same time, converged data is associated with the ring number as an index, the data is classified and organized, and further classified according to the station area information to which the ring number belongs, preparing for subsequent data display and improving data query speed.
[0037] The data terminal displays the converged data after classification, mileage positioning information, and ring number association results in real time, in the form of tables and 3D models. Staff can query, modify, and delete the current detection data through the analysis terminal, which facilitates subsequent data management and tunnel maintenance decisions.
[0038] like Figure 3 The table shown is a statistical table for the accuracy verification of this embodiment. It includes multiple verification points, each containing multiple verification plates, and both forward and reverse verifications were performed on different axes. The verification results show that all accuracy requirements are met.
[0039] In another embodiment of the present invention, a circle center positioning accuracy optimization step is performed on the filtered and denoised point cloud data, including: 1-1 Prepare the corresponding 3D point cloud data; 1-2 Convert the 3D point cloud data into a 2D array through planar fitting; 1-3 Convert the 3D point cloud data into grayscale data to prepare for subsequent algorithms.
[0040] 2-1 Construct a basic variance matrix using a two-dimensional array; 2-2 Construct a plane equation based on the basic variance matrix to eliminate most noise; 2-3 Obtain new profile point data to form a standard two-dimensional array; 2-4 Apply the standard two-dimensional array and use the least squares method to fit the data to a circle; 2-5 Copy the current profile point data and sort it by angle to prepare for subsequent algorithms; 2-6 Solve the linear equation system based on the fitted circle to obtain the center coordinates A1(X,Y,R); 2-7 Transform the A1 coordinates to a 3D coordinate system to obtain point A(X,Y,R); 2-8 Store the center coordinates A in a cache. 3-1 Obtain the profile point data from point 2-5, and use the chord-tangent method to select data points at 10-degree intervals and connect them to form chords (P1, P2, ...); 3-2 Take the midpoint of each chord (P1, P2, ...) and draw its perpendicular bisector to obtain a set of continuous straight lines; 3-3 Connect two adjacent perpendicular bisectors and use their intersection point (H1, H2, ...) as the basic data to construct a system of linear equations; 3-4 Calculate the average of the coordinates of all intersection points (H1, H2, ...) as the center coordinates, and calculate the average distance of all points to the center as the radius to obtain point B1(X, Y, R); 3-5 Transform the coordinates of point B1 to a 3D coordinate system to obtain point B(X, Y, R); 3-6 Store the center coordinates of point B in a cache. 4-1 Using the grayscale image transformed in steps 1-3 as a basis, extract its outline data; 4-2 Use a denoising algorithm to eliminate noise in the grayscale image; 4-3 Apply the Hough algorithm to the denoised data; 4-4 After calculation, obtain the center coordinates C1(X,Y,R); 4-5 Transform the C1 coordinates to a 3D coordinate system to obtain point C(X,Y,R); 4-6 Store the center coordinate point C in a cache; 5-1 Based on the coordinates of the center of the three points A, B, and C obtained in steps 2-8, 3-6, and 4-6, a comprehensive algorithm is used to check whether the three points are collinear. 5-2 If the three points are not collinear, connect the three points to form a triangle. The center of its circumcircle is the final center O(X,Y,R). The radius is calculated as follows: first, calculate the average distance and standard deviation from the center to all points; then, based on the average radius and standard deviation, remove discrete data points; finally, recalculate the average radius based on the remaining data points to obtain the final R. 5-3 If the three points are collinear, the midpoint of the line connecting the three points is taken as the center O(X,Y,R). The radius is calculated in the same way as in 5-2: calculate the average distance and standard deviation from the center to all points; then, based on the average radius and standard deviation, remove discrete data points; finally, recalculate the average radius based on the remaining data points to obtain the final R. 5-4 Process completed.
[0041] like Figure 4The table shown is a statistical table of algorithm deviations in this embodiment. Through individual and comprehensive statistics, it is shown that the deviation of this method is kept to a small level when combined with the least squares method, the tangent method, the Hough algorithm and the comprehensive algorithm.
[0042] In another embodiment, the 3D cameras are specifically defined as six Ranger3 cameras, such as... Figure 5 As shown, a specific installation method is given: six Ranger3 cameras are installed on the surface of the detection vehicle, forming an irregular circle. The distance and angle between adjacent Ranger3 cameras can be adjusted according to the actual situation to ensure coverage of the entire detection range.
[0043] In other embodiments, the 3D camera may also be mounted in a regular circle, or two symmetrical semicircles, etc.
[0044] In other embodiments, a greater number of Ranger3 cameras may be selected as the 3D camera to further ensure detection accuracy.
[0045] like Figure 1 As shown, in one embodiment, the present invention provides a tunnel convergence deformation detection method based on a 3D camera, comprising the following steps: S1: Real-time acquisition of point cloud data of the tunnel using multiple 3D cameras installed on the inspection vehicle; S2: Obtain the ring number information of the tunnel in real time, and associate the point cloud data with the ring number information; S3: Calculate the deflection angle of the detected vehicle in real time, and correct the deflection angle of the detected vehicle. S4: Repeat steps S1 to S3 to stitch together all the obtained point cloud data; S5: Filter and denoise the point cloud data obtained in S4; S6: Optimize the center positioning accuracy of the point cloud data obtained in S5; S7: Perform feature recognition on the point cloud data obtained in S6 to obtain the structural components of the tunnel.
[0046] According to a preferred embodiment, S1 includes: S11: Determine the data acquisition sequence of multiple 3D cameras on the inspection vehicle; S12: Detect vehicle movement and collect point cloud data of the tunnel sequentially based on the collected data; S13: Convert the point cloud data into a grayscale image.
[0047] According to a preferred embodiment, S2 includes: S21: Obtain the ring number information of the tunnel based on the grayscale image; S22: Associate the point cloud data and the ring number information according to the ring number information.
[0048] According to a preferred embodiment, S3 includes: S31: Preset the normal travel angle of the detection vehicle and calculate the deflection angle of the detection vehicle in real time; S32: Based on the normal travel angle and deflection angle, correct the deviation of the detection vehicle.
[0049] According to a preferred embodiment, S4 includes: S41: Pre-calibrate the travel distance of the detection vehicle, and repeat steps S1 to S3 continuously until the travel distance is reached; S42: When the travel distance is reached, a stop message is sent to the detection vehicle; S43: Stitch together all the obtained point cloud data.
[0050] According to a preferred embodiment, S5 includes: S51: Perform processing to remove abnormal light spots, dust, and noise; S52: Perform smoothing filtering.
[0051] According to a preferred embodiment, S6 includes: S61: Based on the point cloud data obtained in S5, obtain the corresponding two-dimensional array and grayscale image data; S62: Construct a basic variance matrix based on the two-dimensional array, construct a plane equation based on the variance matrix, perform circle fitting on the data using the least squares method to obtain the first profile point data, solve the plane equation based on the profile point data, and obtain the coordinates of the first circle center. S63: Based on the first profile point data, select data points at certain angular intervals and connect them to form multiple chords. Draw the perpendicular bisector of each chord according to its midpoint to obtain multiple sets of continuous straight lines. Connect two adjacent perpendicular bisectors and use their intersection as the basic data to construct a system of linear equations. Calculate the average value of the coordinates of all intersection points as the center coordinates of the circle, and calculate the average value of the coordinates of all points to the center coordinates of the circle as the radius to obtain the second center coordinates of the circle. S64: Based on the grayscale image data, obtain the second profile point data, and apply the Hough circle transform algorithm to the second profile point data to obtain the coordinates of the third circle center; S65: Check if the coordinates of the first, second, and third center points are collinear; if they are not collinear, connect the three points to form a triangle, and the center of its circumcircle is the fourth center; if the three points are collinear, take the midpoint of the line connecting the three points as the fourth center; calculate the average distance and standard deviation from the fourth center to all points, remove discrete data points based on the average radius and standard deviation, and then recalculate the average radius based on the remaining data points to obtain the final center. S66: Optimize the center positioning accuracy based on the final center.
[0052] According to a preferred embodiment, S7 includes: S71: Perform feature recognition and segmentation on the point cloud data obtained in S6; S72: Obtain the structural components of the tunnel and perform structural testing on the structural components.
[0053] According to a preferred embodiment, the structural detection includes ellipticity detection, convergence detection, and misalignment detection.
[0054] According to the above embodiments, the tunnel convergence deformation detection method and system based on 3D cameras provided by the present invention collects tunnel point cloud data in real time by setting up multiple 3D cameras on the detection vehicle. This eliminates the need for manual on-track equipment deployment and repeated calibration, solving the problems of cumbersome manual intervention and debugging steps required by traditional lidar detection. Simultaneously, the detection can proceed with the vehicle's movement, allowing for rapid long-distance line detection as needed, thus addressing the issue of low detection efficiency. Collaborative acquisition by multiple 3D cameras generates denser raw point cloud data. Combined with the point cloud stitching step, it eliminates the need for multiple detections or speed reductions to ensure data volume, solving the problem of low point cloud generation density. Furthermore, this method effectively compensates for the impact of driving deviations on the detection results by calculating and correcting the vehicle's deflection angle in real time, combined with optimization of the center positioning accuracy. This breaks through the 5-30mm accuracy limitation of traditional lidar, bringing it closer to the ±5mm tunnel convergence detection standard, solving the problem of limited detection accuracy. Overall, it possesses advantages such as high automation, high detection efficiency, excellent point cloud data quality, and detection accuracy that better meets standards, enabling more accurate monitoring of tunnel convergence status and providing more reliable protection for the safe operation of subways.
[0055] The embodiments of this application have been described in detail above. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.
Claims
1. A method for detecting tunnel convergence deformation based on a 3D camera, characterized in that, Includes the following steps: S1: Real-time acquisition of point cloud data of the tunnel using multiple 3D cameras installed on the inspection vehicle; S2: Obtain the ring number information of the tunnel in real time, and associate the point cloud data with the ring number information; S3: Calculate the deflection angle of the detected vehicle in real time, and correct the deflection angle of the detected vehicle. S4: Repeat steps S1 to S3 to stitch together all the obtained point cloud data; S5: Filter and denoise the point cloud data obtained in S4; S6: Optimize the center positioning accuracy of the point cloud data obtained in S5; S7: Perform feature recognition on the point cloud data obtained in S6 to obtain the structural components of the tunnel.
2. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S1 includes: S11: Determine the data acquisition sequence of multiple 3D cameras on the inspection vehicle; S12: Detect vehicle movement and collect point cloud data of the tunnel sequentially based on the collected data; S13: Convert the point cloud data into a grayscale image.
3. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S2 includes: S21: Obtain the ring number information of the tunnel based on the grayscale image; S22: Associate the point cloud data and the ring number information according to the ring number information.
4. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S3 includes: S31: Preset the normal travel angle of the detection vehicle and calculate the deflection angle of the detection vehicle in real time; S32: Based on the normal travel angle and deflection angle, correct the deviation of the detection vehicle.
5. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S4 includes: S41: Pre-calibrate the travel distance of the detection vehicle, and repeat steps S1 to S3 continuously until the travel distance is reached; S42: When the travel distance is reached, a stop message is sent to the detection vehicle; S43: Stitch together all the obtained point cloud data.
6. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S5 includes: S51: Perform processing to remove abnormal light spots, dust, and noise; S52: Perform smoothing filtering.
7. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S6 includes: S61: Based on the point cloud data obtained in S5, obtain the corresponding two-dimensional array and grayscale image data; S62: Based on the two-dimensional array structure, the least squares method is used to fit the data to a circle to obtain the first profile point data. Based on the profile point data, the plane equation is solved to obtain the coordinates of the first circle center. S63: Based on the first profile point data, the coordinates of the second circle center are obtained using the tangent method; S64: Based on the grayscale image data, the Hough algorithm is used to obtain the coordinates of the third circle center; S65: Check if the coordinates of the first, second, and third center points are collinear; if they are not collinear, connect the three points to form a triangle, and the center of its circumcircle is the fourth center; if the three points are collinear, take the midpoint of the line connecting the three points as the fourth center; calculate the average distance and standard deviation from the fourth center to all points, remove discrete data points based on the average radius and standard deviation, and then recalculate the average radius based on the remaining data points to obtain the final center. S66: Optimize the center positioning accuracy based on the final center.
8. The tunnel convergence deformation detection method based on a 3D camera according to claim 1, characterized in that, S7 includes: S71: Perform feature recognition and segmentation on the point cloud data obtained in S6; S72: Obtain the structural components of the tunnel and perform structural testing on the structural components.
9. The tunnel convergence deformation detection method based on a 3D camera according to claim 8, characterized in that, In step S7, the structural detection includes ellipticity detection, convergence detection, and misalignment detection.
10. A tunnel convergence deformation detection system based on a 3D camera, characterized in that, include: The hardware system includes: a 3D camera, photoelectric encoder, and vehicle posture compensation device installed on the detection vehicle, and multiple electronic tags installed at fixed distances in the tunnel; The 3D camera is used to collect point cloud data of the tunnel; The photoelectric encoder is fixedly installed on the axle of the detection vehicle and is used to calculate the mileage and speed, while triggering the 3D camera at a fixed frequency. The vehicle posture compensation device is located in the middle of the detection vehicle to dynamically compensate for the vehicle body deflection angle during the movement. Adjacent electronic tags maintain a fixed distance to assist in obtaining tunnel ring number information; Software equipment systems, including data collection systems and data analysis systems; The data collection system is responsible for collecting point cloud data from the 3D camera, mileage and speed data from the photoelectric encoder, real-time vehicle attitude data provided by the vehicle attitude compensation device, and corresponding electronic tag data in real time. The data analysis system is used to analyze the data collected by the data collection system to achieve three-dimensional detection of the tunnel.