Roadway section convergence multi-view three-dimensional reconstruction method and monitoring system thereof

Abstract

The invention relates to the technical field of mine safety monitoring, in particular to a roadway section convergence multi-view three-dimensional reconstruction method and a monitoring system thereof.The method comprises the steps that a rectangular plane coordinate system is established on a roadway section, a plurality of measuring areas are arranged in the circumferential direction, a plurality of equal-height measuring points are evenly distributed in each measuring area in the elevation direction of the section, and every two adjacent measuring areas share at least one measuring point; a laser structured light projector and a CMOS receiving camera are arranged at each measuring point; matching the feature points through an improved stereo matching algorithm to obtain depth feature point information and establish a section point cloud; dividing deformation into local deformation and overall deformation, establishing deformation association between measurement points through deformation feature description, and obtaining overall convergence information of the section by adopting a point cloud registration method after local deformation compensation is formed; a section safety evaluation model is established, section deformation evaluation and reinforcement suggestions are given, all-directional non-blind area monitoring of the section is achieved, and the monitoring precision reaches the millimeter level.

Description

Technical Field

[0001] The present invention relates to the field of mine safety monitoring technology, and specifically to a multi-perspective three-dimensional reconstruction method for tunnel section convergence and a monitoring system thereof, which are used to accurately monitor the deformation convergence of mine tunnel sections and provide early warning and support decision support for tunnel safety. Background Art

[0002] As mining extends deeper, ground pressure activity becomes increasingly frequent and intense, leading to increasingly severe roadway deformation, threatening mine safety. Traditional methods for monitoring roadway cross-section convergence primarily rely on single-point measurement equipment such as inclinometers and convergence meters. These methods suffer from large blind spots, insufficient accuracy, and an inability to reflect overall deformation trends, making them inadequate for deep mine safety monitoring.

[0003] While existing optical monitoring technologies enable non-contact measurement, they often only observe from a single perspective and cannot fully capture cross-sectional deformation information. While 3D laser scanning technology can acquire high-precision point cloud data, the equipment is expensive, complex to operate, and difficult to achieve long-term continuous monitoring. Furthermore, existing methods for deformation detection and analysis generally lack a mechanism for collaboratively identifying local and global deformation, resulting in an inadequate understanding of deformation transmission patterns, which impacts early warning effectiveness and decision-making support capabilities.

[0004] Therefore, there is an urgent need for a comprehensive, high-precision, and intelligent tunnel section convergence monitoring method to provide timely warning of tunnel deformation and effective support decision-making support. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-perspective 3D reconstruction method for tunnel cross-section convergence and a monitoring system thereof. By means of overlapping layout of multiple measurement areas, an improved stereo matching algorithm, and collaborative identification of local and overall deformations, all-round, high-precision, and intelligent monitoring of tunnel cross-sections can be achieved, providing timely warnings and effective decision-making support for tunnel safety.

[0006] The present invention proposes a tunnel cross-section convergence multi-view 3D reconstruction method, comprising:

[0007] A plane rectangular coordinate system is established on the tunnel cross section, and multiple measurement areas are arranged around the cross section. Each measurement area has a number of equal-height measurement points evenly distributed along the cross section elevation direction. At least one measurement point is shared by both adjacent measurement areas. A laser structured light projector is arranged at each measurement point. The normal direction of the laser structured light shadow surface is perpendicular to the tunnel contour line. A CMOS receiving camera is set next to each laser structured light projector. A threshold S2 is set for the overlapping portion of the shooting area of ​​each laser structured light projector and the adjacent CMOS camera. Multiple laser structured light groups are set around the cross section. Each laser structured light group includes a structured light projector and a CMOS receiving camera. The number of laser structured light groups is determined by the cross section size.

[0008] Based on the laser beam of the projector, surface feature projection encoding is performed on each measuring point and a preset measuring area circumferentially of the measuring point. The CMOS receiving camera receives the projection echo and extracts feature points of the measuring point and the measuring area circumferentially thereof. The feature points of each measuring point and the measuring area circumferentially thereof are matched using an improved stereo matching algorithm to obtain depth feature point information of the measuring point in the observation area, establish cross-sectional point cloud data, and calculate cross-sectional convergence information.

[0009] The cross-section convergence deformation is divided into local deformation and global deformation. The local deformation refers to the convergence deformation within the current measurement area, and the global deformation includes the convergence deformation transfer between the current measurement area and the adjacent measurement area. In the cross-section local deformation detection stage, each measurement point is used as a feature, and feature matching is performed within the observation area of ​​the cross-section laser structured light group and the CMOS camera group to obtain the cross-section convergence information of each measurement area. In the cross-section global deformation detection stage, the cross-section convergence information of the current measurement area is obtained, and at the same time, the deformation features of the adjacent measurement points are described according to the threshold S2. The deformation association between the measurement points is established based on the deformation feature description. After forming local deformation compensation, the point cloud registration method is used to perform cross-section matching to obtain the cross-section global convergence information.

[0010] Based on current convergence information, historical convergence information, and real-time environmental monitoring information, a section safety evaluation model is established to obtain a section deformation assessment, issue a warning for abnormal section deformation, and display real-time monitoring data and section convergence trend curves, and provide reinforcement recommendations for subsequent section deformation.

[0011] Preferably, the measurement area preset in the circumferential direction of the measurement point is within a range with the measurement point as the center and R as the radius.

[0012] Preferably, the measuring points are arranged in sequence along the elevation direction of the section.

[0013] Preferably, at least one measuring point between two adjacent measuring areas is shared by both areas, which means that at least one pair of adjacent edges between the measuring areas overlap, the overlapping length of the adjacent edges is not less than R, and the area where the measuring points overlap is within the overall cross-sectional contour.

[0014] Preferably, the improved stereo matching algorithm establishes a mathematical model to achieve accurate matching of feature points and obtain depth feature point information by combining parameters such as parallax value, laser beam cross-sectional area, camera intrinsic parameters, camera focal length, camera spacing, lens focal length, and the distance between the camera reference point and the laser structure light source point.

[0015] Preferably, during the overall deformation detection, the cross-sectional contour data of adjacent measurement areas are divided according to the deformation characteristics of different measurement points into contour data of different measurement points, and the deformation data of the measurement points of the current measurement area are matched with the deformation data of the measurement points of the adjacent measurement areas to complete the deformation point matching of the current measurement point to the adjacent position measurement area. Through the matching of the measurement point deformation data, the constraint relationship between the cross-sectional deformation points is established, the overall deformation of the cross section is established, and the overall convergence information of the cross section is obtained.

[0016] Preferably, the preset measurement areas at the measurement points include two types, one is the circumferential measurement area of ​​the measurement points on the cross section, and the other is the measurement area between measurement areas and between measurement points on the cross section, and the two measurement areas partially overlap.

[0017] Preferably, when the cross-section state and deformation information is output, the measurement points of the cross-section, the distribution density of the measurement points, the average convergence of the measurement points, the maximum convergence of the measurement points, the convergence change ratio of the measurement points, the convergence trend of the measurement points, the overall convergence of the cross-section, the convergence curve, the convergence distribution diagram, the safety status of the measurement points, and the safety status of the cross-section state are used as parameters to display the real-time monitoring data and the cross-section convergence trend curve, give reinforcement suggestions for subsequent cross-section deformation, establish a cross-section safety evaluation model, and display the cross-section safety hazards in a graded manner based on the established cross-section safety evaluation model.

[0018] Preferably, the deformation feature description includes deformation amount, deformation direction, deformation rate and deformation distribution, and a deformation association network between measurement points is constructed through the deformation feature description; the deformation association network includes direct association, indirect association, forced association and time series association, which is used to analyze the transmission law of deformation in space and assist in deformation trend prediction.

[0019] The tunnel cross-section convergence multi-view 3D reconstruction monitoring system includes a measurement area information control module, a laser projection module for each measurement point, a circumferential measurement module for the measurement point, a deformation matching module for the measurement point, a laser structured light projector and a CMOS sensor group, a measurement area contour registration module, a cross-section convergence information calculation module, a cross-section convergence trend and status evaluation module, a cross-section display module, and a cross-section deformation feature description module. The laser projection module for each measurement point, the circumferential measurement module for the measurement point, the cross-section convergence information calculation module, the cross-section convergence trend and status evaluation module, and the cross-section display module are connected in sequence. The measurement area information control module, the cross-section convergence information calculation module, the cross-section display module, and the laser projection modules of each measurement point are connected in sequence; the laser structured light projector and the CMOS sensor group are connected to the multiple input ends of the cross-section convergence information calculation module, the cross-section convergence trend and state evaluation module, and the cross-section display module; the multiple input ends of the laser projection modules of each measurement point, the measurement point circumferential measurement module, the measurement area contour registration module, and the cross-section convergence information calculation module are connected to each other; the cross-section convergence information calculation module is connected to the cross-section three-dimensional display module; the laser projection modules of each measurement point are connected to the cross-section convergence information calculation module; the laser projection modules of each measurement point are ... The output terminals of the block, the measuring point circumferential measurement module, the measuring point deformation matching module, and the measuring area contour registration module are connected to each other; the measuring area information control module is used to obtain the circumferential information of the cross section, the overlapping information between adjacent measuring areas, the constraint relationship between the measuring areas, and the control of the measuring area information; the laser projection modules of each measuring point are used to generate coded projection signals to realize the projection coding of the surface features of the measuring area; the measuring point circumferential measurement module is used to collect feature information of each measuring point on the cross section and the measuring area preset in the circumference of the measuring point; the measuring point deformation matching module, Used to establish feature information for each measurement point and realize feature matching of each measurement point in the section; the laser structured light projector and CMOS sensor group are used to collect the cross-sectional contours of each measurement area and the measurement results of the measurement points; the measurement area contour alignment module is used to establish the constraint relationship between the cross-sectional measurement points; the cross-sectional convergence information calculation module is used to match the feature points of the measurement area and calculate the cross-sectional convergence information; the cross-sectional convergence trend and status assessment module is used to establish a cross-sectional safety assessment model; the cross-sectional display module is used to display real-time monitoring data and the cross-sectional convergence trend curve.

[0020] The present invention has the following beneficial effects:

[0021] 1. Through the overlapping layout of multiple measurement areas and the heterogeneous data acquisition structure, 360° blind-spot monitoring of the cross section is achieved, and the monitoring accuracy is improved from the traditional centimeter level to the millimeter level, significantly improving the comprehensiveness and accuracy of monitoring.

[0022] 2. The improved stereo matching algorithm comprehensively considers multi-dimensional factors such as the laser beam cross-sectional area and camera parameters, which increases the feature point matching accuracy by more than 30%, significantly improving the accuracy of 3D reconstruction.

[0023] 3. An innovative local and global deformation classification and processing mechanism is proposed to achieve multi-scale decoupling and fusion of deformation, which can simultaneously capture local subtle changes and overall deformation trends, and advance the warning time by 30% to 50%.

[0024] 4. The deformation feature description and measurement point association construction technology makes deformation transfer recognition more accurate. The multi-source data fusion driven by point cloud registration improves the overall perception capability, and the system reliability reaches more than 99.5%.

[0025] 5. The comprehensive safety assessment model can not only issue early warnings in a timely manner, but also provide targeted support and reinforcement recommendations, achieving a shift from passive monitoring to active prevention, and reducing maintenance costs by 20% to 40%. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the tunnel cross-section convergence multi-view 3D reconstruction method of the present invention;

[0027] Figure 2 This is a schematic diagram of the multi-survey area layout of the present invention;

[0028] Figure 3 This is a module composition diagram of the tunnel section convergence monitoring system of the present invention;

[0029] Figure 4 This is a flowchart of the collaborative identification of local and overall deformations of the present invention;

[0030] Figure 5 Schematic diagram of the cross-section safety assessment model of the present invention. DETAILED DESCRIPTION

[0031] Please refer to the attached Figure 1-5 , the specific implementation of the present invention is further described in detail below with reference to the accompanying drawings.

[0032] like Figure 1 As shown, the tunnel cross-section convergence multi-view 3D reconstruction method provided by the present invention includes the following steps:

[0033] 1. Multi-survey area overlapping layout and data acquisition structure:

[0034] like Figure 2As shown, a rectangular coordinate system is established on the tunnel cross section, and multiple measurement zones are arranged around the cross section. Each measurement zone has several equal-height measurement points evenly distributed along the cross section elevation. The number of measurement points is determined by the cross section dimensions and monitoring accuracy requirements, typically ranging from 4 to 12. For example, a typical horseshoe-shaped tunnel cross section with a width of 5 meters and a height of 4 meters can be arranged with 6 measurement zones, each with 8 measurement points, for a total of 48 measurement points, ensuring full cross section coverage and sufficient data redundancy. At least one measurement point between two adjacent measurement zones is shared by both to ensure data continuity.

[0035] A laser structured light projector is deployed at each measurement point, with the normal of the laser structured light shadow plane perpendicular to the roadway contour for optimal projection. For example, for measurement points on the roadway roof, the projector is oriented vertically downward; for measurement points on the sidewall, the projector is oriented perpendicular to the sidewall. A CMOS receiving camera is located adjacent to each laser structured light projector to receive the structured light projection echo.

[0036] A threshold S2 is set for the overlap between each laser structured light projector and the adjacent CMOS camera's capture area. This threshold is typically set between 30% and 50%. In practical applications, such as in a cross-section monitoring system for a coal mine excavation face, setting S2 to 40% ensures sufficient overlap for data fusion while avoiding redundancy caused by excessive overlap. Multiple laser structured light groups are set along the circumference of the cross section. Each group consists of a structured light projector and a CMOS receiving camera. The number of laser structured light groups is determined by the cross-section size to ensure full cross-section coverage.

[0037] Preferably, the measurement area pre-set around the measurement point is within a radius R, centered at the measurement point. R is typically 5% to 10% of the tunnel radius. For example, in a metal mine haulage tunnel with a cross-sectional radius of 3 meters, R can be set to 20 centimeters to ensure adequate coverage and overlap. The measurement points are arranged sequentially along the cross-sectional elevation direction, forming an orderly monitoring network.

[0038] In one embodiment of the present invention, at least one measurement point between two adjacent measurement areas is shared by both areas, which means that at least one pair of adjacent edges between the measurement areas overlap, the overlapping length of the adjacent edges is not less than R, and the area where the measurement points overlap is within the overall cross-sectional contour. Figure 2 Taking the hexagonal survey area layout shown as an example, adjacent survey areas share a side with a length of about 50 cm, and the shared measurement points are located at the end points and midpoints of the side. This design ensures the continuity and consistency of data between survey areas, laying the foundation for subsequent overall deformation analysis.

[0039] There are two types of preset measurement areas at each measurement point: one is the circumferential measurement area around the measurement point on the cross section, and the other is the measurement area between measurement areas, that is, between measurement points on the cross section. The two measurement areas partially overlap. In practical applications, such as in a deep soft rock tunnel, the circumferential measurement area covers a 20-centimeter radius around the measurement point, while the inter-measurement area covers ±15 centimeters around the measurement area boundary. The two overlap by approximately 10 centimeters at the boundary. This dual-layer measurement area design enhances the robustness and comprehensiveness of data acquisition.

[0040] 2. Surface feature extraction and stereo matching:

[0041] The projector's laser beam projects surface features onto each measurement point and the pre-defined measurement area around it. In practical applications, such as when using a combination of shotcrete and anchor bolts for roadway support, a random speckle encoding pattern with a density of 8 coding units per square centimeter can be used. This encoding method effectively addresses the low-texture characteristics of shotcrete surfaces and improves feature matching success rates.

[0042] The CMOS receiving camera receives the projected echo and extracts feature points at the measurement point and its surrounding area. This feature extraction utilizes a multi-scale feature extraction strategy, extracting large-scale structural features first and then small-scale detail features from a coarse-to-fine approach. For example, in the main tunnel of a metal mine, feature points are hierarchically divided into primary feature points (support structure points such as anchor caps), secondary feature points (cross-section edge points), and tertiary feature points (surface texture points). One, 30, and 200 feature points are extracted, respectively, to form a complete feature hierarchy.

[0043] An improved stereo matching algorithm is used to match the feature points of each measurement point and its circumferential measurement area. This algorithm considers multi-dimensional parameters such as parallax, laser beam cross-sectional area, camera intrinsic parameters, camera focal length, camera spacing, lens focal length, and the distance between the camera reference point and the laser structure light source, and establishes a mathematical model to achieve accurate matching of feature points.

[0044] The improved stereo matching algorithm can be expressed as:

[0045]

[0046] in: is the matching cost function, dimensionless; is the pixel position in pixels; is the disparity value in pixels; and is the grayscale value of the corresponding point, ranging from 0 to 255, dimensionless; is the weighting function considering the cross-sectional area of ​​the laser beam, dimensionless; is the normalization coefficient, which makes the magnitude of the grayscale term match the smoothing term, and its unit is the unit of the smoothing term / the square unit of the grayscale difference; is the smoothing coefficient, dimensionless; is a smoothing term and is dimensionless.

[0047] The weighting function can be further expressed as:

[0048]

[0049] in: is the cross-sectional area of ​​the laser beam in square millimeters; the function value has been converted to dimensionless after exponential operation.

[0050] The smoothness term can be expressed as:

[0051] ,

[0052] in: The truncation threshold is typically set to 5 pixels to prevent outliers from excessively influencing the results. In an actual coal mine gas drainage tunnel monitoring system, the matching process includes five steps: image preprocessing (grayscale conversion and Gaussian filtering), preliminary feature matching (based on SIFT descriptors), matching optimization (elimination of outlier matching pairs), depth calculation, and 3D reconstruction of feature points. The matching window size is set to 11×11 pixels, the disparity search range is ±30 pixels, and the matching confidence threshold is set to 0.75. After field testing, these parameters can achieve a matching success rate of over 95% in the complex environment of the tunnel.

[0053] The depth calculation formula is:

[0054] ,

[0055] in: is the depth value in millimeters; is the focal length in millimeters; is the baseline length in millimeters; is the disparity value in pixels; is the sensor pixel pitch, in millimeters / pixel. , which ensures the consistency of units and makes the final calculation result millimeters.

[0056] Through this process, we obtain the depth feature points of the measurement points within the observation area, create cross-sectional point cloud data, and calculate cross-sectional convergence information. For example, in an underground iron ore haulage tunnel, the point cloud data density reached 120 points per square centimeter, with a coverage rate of 97% and a point position accuracy of better than 1.5 mm, meeting the requirements of high-precision monitoring.

[0057] 3. Local-global dual-scale deformation decoupling and identification:

[0058] This paper innovatively divides cross-section convergence deformation into two categories: local deformation and global deformation. Local deformation refers to convergence deformation within the current measurement area, focusing on microscopic changes; global deformation includes the transfer of convergence deformation between the current measurement area and adjacent measurement areas, focusing on macroscopic trends.

[0059] During the local deformation detection phase, feature matching is performed within the observation areas of the laser structured light and CMOS camera groups, using each measurement point as a feature. This allows for convergence information to be obtained for each measurement area. For example, the local deformation detection process for a coal mine's haulage lane monitoring system includes five steps: data acquisition (every 10 minutes), surface feature extraction (50 feature points), feature matching (matching success rate >90%), displacement calculation, and deformation determination. The displacement threshold is set at 2 mm. The system only confirms local deformation when the displacement exceeds 2 mm for two consecutive monitoring measurements and the consistency of the displacement vector direction is greater than 85%. This design effectively avoids misjudgments caused by equipment jitter or temporary interference.

[0060] In the overall deformation detection stage of the section, after obtaining the convergence information of the section in the current measurement area, the deformation characteristics of the adjacent measurement points are described according to the threshold S2. The deformation association between the measurement points is established based on the deformation feature description. After local deformation compensation is formed, the point cloud registration method is used to perform section matching to obtain the overall convergence information of the section.

[0061] The deformation feature description includes four elements: deformation amount, deformation direction, deformation rate and deformation distribution. The deformation feature descriptor can be expressed as:

[0062] ,

[0063] in: is the deformation feature descriptor, a dimensionless vector; is the deformation variable, the unit is millimeter, indicating the displacement; is the deformation direction vector, a dimensionless unit vector, indicating the displacement direction; is the deformation rate, in mm / day, which indicates the deformation per unit time; It is the deformation distribution feature, which can be expressed as the gradient vector of the deformation in space, with the unit of mm / m

[0064] In a deep mining tunnel of a metal mine, when the roof of a certain measurement area sinks by 3.5 mm, a specific deformation feature descriptor is formed: deformation variable M = 3.5 mm, deformation direction Vertically downward, the deformation rate R = 0.7 mm / day (average value for 5 consecutive days), and the deformation distribution P shows a bell-shaped distribution with the largest sinking in the center and decreasing toward the surrounding areas, with a gradient of about 0.15 mm / cm.

[0065] The deformation association network between measurement points is constructed by describing the deformation features. This network includes four types of associations: direct association, indirect association, forced association, and temporal association. The association strength is calculated using the weighted sum of spatial distance and deformation similarity:

[0066] ,

[0067] in: is the strength of the association between the measurement point and j, dimensionless, ranging from [0,1]; sim is the similarity of the deformation feature descriptor, dimensionless, and its value range is is the spatial distance between measurement points, in meters; is the characteristic distance parameter, in meters, which controls the attenuation rate of the distance effect; is the weight coefficient, dimensionless, ranging from [0,1]; Convert the distance to a dimensionless attenuation factor in the range (0,1). The similarity calculation formula remains the same, but make sure all components are dimensionless:

[0068] ,

[0069] in: is the weight coefficient, satisfying Each similarity function They are the similarity calculation functions for shape variable, direction, rate and distribution, respectively. They are all designed to be dimensionless and have a value range of [0,1].

[0070] In the tunnel monitoring system above a certain iron mine, Set to 0.6, , forming an association strategy that emphasizes deformation direction and deformation amount. Practice has proved that this setting is most effective in capturing the tunnel deformation transmission caused by stope activities.

[0071] The deformation association network is used to analyze the spatial transmission law of deformation and assist in deformation trend prediction. The threshold for deformation transmission is set as the correlation strength. When this threshold is exceeded, deformation is considered to be effectively transferred between measurement points i and j. In a large-section coal roadway example, the deformation association network successfully predicted a roof collapse caused by roof delamination, bringing the warning time forward by approximately 40 hours, buying valuable time for evacuation and support reinforcement.

[0072] When performing global deformation detection, the cross-sectional profile data of adjacent measurement areas is divided into profile data for different measurement points based on the deformation characteristics of the different measurement points. The deformation data of the current measurement point is then matched with the deformation data of the adjacent measurement points to complete the deformation point matching between the current measurement point and the adjacent measurement area. By matching the deformation data of the measurement points, constraints are established between the cross-sectional deformation points, forming a global deformation model for the cross section, and ultimately obtaining global convergence information for the cross section.

[0073] Point cloud registration uses the improved iterative closest point (ICP) algorithm, which can be expressed as:

[0074] ,

[0075] in: is the registration error function, dimensionless, the smaller the value, the better the registration effect; is the rotation matrix, a 3×3 dimensional matrix, representing the point cloud rotation transformation; is the translation vector, a 3D vector in millimeters, representing the translation transformation of the point cloud; and are corresponding point pairs, each is a 3D vector, in millimeters; is the point pair weight, dimensionless, ranging from [0,1], and related to the deformation feature; is the number of point pairs, usually 1000-10000 point pairs; is the regularization coefficient, usually 0.01-0.1, dimensionless; To consider the regularization term of deformation characteristics, dimensionless

[0076] The regularization term can be expressed as:

[0077] ,

[0078] in: is the regularization term, dimensionless; is the rotation matrix, 3×3 dimensions, dimensionless; is a 3×3 identity matrix, dimensionless; is the Frobenius norm, which indicates the degree of rotation deviation and is dimensionless; is the translation vector in millimeters; is the Euclidean norm of the translation vector, in millimeters; is the characteristic length of the system, in millimeters (can be set as the characteristic size of the tunnel section); is the weight coefficient, dimensionless, usually taken as 0.7 and 0.3 respectively. , so that the second Converted to dimensionless, so that the entire regularization term maintains dimension consistency.

[0079] In practical applications, these modified algorithm formulas have been verified in multiple mining environments. For example, in a certain potash mine horizontal transport tunnel monitoring system, the modified point cloud registration algorithm was used, and the registration accuracy reached times, fully meeting the actual needs of high-precision deformation monitoring.

[0080] In a horizontal transport tunnel monitoring system of a potash mine, this algorithm is used for registration. The iteration termination condition is that the error change is less than 0.008 mm or the maximum number of iterations (set to 40 times) is reached. The average registration time is 1.2 seconds per time, and the registration accuracy reaches ±0.5 mm, meeting the needs of high-precision deformation monitoring.

[0081] 4. Safety evaluation and decision support:

[0082] Based on the current convergence information, historical convergence information, and real-time environmental monitoring information, a section safety evaluation model is established to obtain the section deformation evaluation results. The evaluation model can be expressed as:

[0083] ,

[0084] in: It is a safety score, dimensionless, ranging from 0 to 100, with higher scores indicating greater safety; is the current convergence information vector, which contains multi-dimensional information such as convergence amount and convergence rate; It is a historical convergence information matrix, which contains the convergence records over the past period of time (usually 7-30 days); It is the environmental monitoring information vector, including auxiliary information such as temperature, humidity, and acoustic emission; is the weight vector, reflecting the importance of each factor; To evaluate the function, methods such as weighted summation or support vector machine are usually used.

[0085] The weighted summation form can be expressed as:

[0086] ,

[0087] in: is the number of evaluation factors, usually 5-10; is the weight of the i-th factor, satisfying ; is the risk score function of the i-th factor, and its value range is [0,100].

[0088] In a safety assessment model for a roadway above a coal mine's mining face, evaluation factors include the current maximum convergence value (weight 0.25), maximum convergence rate (weight 0.3), convergence acceleration (weight 0.2), convergence distribution non-uniformity (weight 0.15), and environmental factors (weight 0.1). When S ≥ 80, the system determines it is in a safe state. When S ≤ S < 80, it enters a caution state, reminding staff to strengthen monitoring. When S ≤ S < 60, it triggers a warning state, recommending local reinforcement measures. When S < 40, the system enters a dangerous state, issuing an evacuation alarm and requiring immediate reinforcement.

[0089] When outputting the section status and deformation information, the parameters include each measurement point in the section, the distribution density of measurement points, the average convergence of measurement points, the maximum convergence of measurement points, the convergence change ratio of measurement points, the convergence trend of measurement points, the overall convergence of the section, the convergence curve, the convergence distribution diagram, the safety status of the measurement points, and the safety status of the section status. The real-time monitoring data and the section convergence trend curve are displayed, and reinforcement suggestions are given for subsequent section deformation.

[0090] During a real-world application in a deep tunnel development project at a metal mine, the system detected a sudden increase in the convergence rate at a roof center measurement point from 0.5 mm / day to 1.8 mm / day, causing the safety score to drop from 85 to 58. The system immediately issued a warning and, based on the deformation characteristics, suggested a potential overhang in the middle of the roof. It recommended reinforcement by adding one central anchor cable and two anchor bolts on each side. The mine promptly implemented the recommended measures, averting a potential collapse.

[0091] The established cross-section safety assessment model displays cross-section safety hazards in a graded manner. On the 3D display interface, safe areas are displayed in green (S ≥ 80), caution areas in yellow (60 ≤ S < 80), warning areas in orange (40 ≤ S < 60), and danger areas in red (S < 40). This intuitive visualization allows staff to identify risk areas at a glance.

[0092] Reinforcement recommendations are based on deformation characteristics and historical experience, using the nearest neighbor algorithm to match the most suitable reinforcement solution from a preset solution library. The nearest neighbor algorithm can be expressed as:

[0093] ,

[0094] in: is the current deformation feature vector; is the feature vector of the i-th historical case; is the feature dimension, usually 5-8 dimensions; is the weight coefficient of the j-th dimension feature; For the distance between the current situation and historical case i, the reinforcement scheme corresponding to the historical case with the smallest distance is selected as the recommended scheme; is the current deformation feature vector; is the feature vector of the i-th historical case; is the feature dimension, usually 5-8 dimensions; is the weight coefficient of the j-th dimension feature; is the distance between the current situation and historical case i.

[0095] Select the reinforcement solution corresponding to the historical case with the smallest distance as the recommended solution:

[0096] ,

[0097] In a typical mine main haulage tunnel monitoring system, the pre-set solution library contains over 50 reinforcement solutions for different deformation types. When a tunnel section is detected exhibiting lateral extrusion deformation (horizontal convergence greater than vertical convergence), the system automatically matches the most similar historical case and recommends increasing the density of sidewall anchor cables and using a combination of prestressed anchor cables and steel belts for roof support, thus guiding on-site support work.

[0098] like Figure 3 As shown, the tunnel cross-section convergence multi-view 3D reconstruction monitoring system provided by the present invention includes the following modules:

[0099] Survey area information control module 1 is used to obtain cross-section circumferential information, overlap information between adjacent survey areas, collect constraints between survey areas, and control survey area information. Specific functions include defining survey area boundaries, planning measurement point distribution, determining overlapping areas, and establishing constraints.

[0100] The boundary of the survey area is usually represented by a parameterized curve in the form of:

[0101] ,

[0102] in: is the boundary curve, a two-dimensional vector function; and is a parametric equation with the unit being meter; It is a parameter variable, dimensionless, and its value range is [0,1].

[0103] In an elliptical tunnel example, the measurement area boundary can be expressed as:

[0104] ,

[0105] ,

[0106] Where a and b are the major and minor axes of the ellipse, respectively, in meters.

[0107] Constraint relationships are represented using a graph structure, with each measurement area as a node, constraints between adjacent measurement areas as edges, and weights reflecting constraint strength. In a large-section haulage roadway monitoring system at a coal mine, six measurement areas are used, forming a ring topology. Each measurement area has constraints with two adjacent areas, and the strength of the constraints is determined by the number of shared measurement points (typically 2-3) and the importance of their locations.

[0108] Laser projection module 2 at each measurement point generates a coded projection signal, enabling projection encoding of surface features in the measurement area. Coding patterns include grids, stripes, or random spots to accommodate varying surface characteristics. In an application involving a pumphouse tunnel in a metal mine, a high-density grid coding pattern (12 coding units per square centimeter) was employed due to the smooth surface of the shotcrete, improving the success rate of texture feature extraction.

[0109] Projection parameters include light intensity (typically 2-4W), angle (perpendicular to the cross-section outline), range (a circular area with a radius of R centered on the measurement point), and resolution (≥1024×768), which are automatically adjusted based on environmental conditions. In high-dust environments underground, the system automatically increases light intensity by approximately 20% to ensure projection clarity.

[0110] Module 3, Circumferential Measurement of Measurement Points, collects characteristic information from each measurement point on the cross section and from a pre-defined measurement area around each measurement point. In a typical coal mine tunneling face rear roadway monitoring system, the acquisition frequency is set at 5 minutes. If a sudden increase in the convergence rate (>1.5 mm / hour) is detected, the frequency is automatically increased to 30 seconds, enabling real-time capture of sudden deformation.

[0111] Data acquisition utilizes hardware triggering, with the main controller sending synchronization signals to each measurement unit to ensure consistent data at each measurement point (time error <10 milliseconds). The acquisition window is set to 100 milliseconds, within which full-section data acquisition is completed. The system utilizes high-speed industrial Ethernet (100Mbps) for data transmission and processing, ensuring real-time data transmission and processing.

[0112] Measurement Point Deformation Matching Module 4 is used to establish feature information for each measurement point and achieve feature matching for each measurement point in the cross-section. In an application in the main haulage tunnel of an iron ore mine, this module uses an improved SURF (Speeded Up Robust Features) algorithm to extract features. 50-80 feature points are extracted from each measurement point area to construct a 128-dimensional feature descriptor. The FLANN (Fast Approximate Nearest Neighbor) algorithm is then used for feature matching, achieving sub-pixel matching accuracy (<0.5 pixel), ensuring high precision for subsequent 3D reconstruction.

[0113] Feature matching results are screened using a geometric consistency check (RANSAC algorithm) to eliminate false matches and retain highly confident matches. Even under complex lighting conditions (such as interference from mining lamps), the matching success rate remains above 85%, demonstrating the robustness of the algorithm.

[0114] The laser structured light projector and CMOS sensor set 5 is used to collect cross-sectional profiles and measurement results at each measurement area. In a coal mine with strict explosion-proof requirements, this intrinsically safe device is configured as follows: a 1280×1024 pixel CMOS camera with a 75-degree field of view and a 35fps frame rate; a 2.5W structured light projector with a wavelength of 780nm (near-infrared to reduce visible light interference) and a projection resolution of 1280×800. All equipment meets the certification requirements for intrinsically safe equipment for mining use.

[0115] The device's housing features an IP67 protection rating, making it resistant to high humidity (relative humidity >95%) and dusty environments. In tunnels with large temperature fluctuations (up to 30°C), the device's built-in temperature compensation algorithm ensures measurement accuracy is unaffected by temperature.

[0116] Module 6, the measurement area contour registration module, is used to establish constraints between cross-sectional measurement points. In a tunnel monitoring system for a potash mine with a thick overburden layer, four types of constraints were identified: spatial proximity constraints (displacement differences between adjacent measurement points < 5 mm), structural consistency constraints (deformation coordination between measurement points within the same support unit > 80%), physical feasibility constraints (deformation must meet the impenetrability principle), and historical trend constraints (continuity between current deformation and historical trends > 70%).

[0117] Constraints were quantified using a weighted graphical model, with weights set to 0.4 for roof measurement points, 0.3 for sidewall measurement points, and 0.3 for floor measurement points. This reflects the dominance of roof behavior during deformation. Through the constraint network, the system successfully identified the overall convergence process caused by roof subsidence, accelerating the early warning time by approximately 36 hours.

[0118] Section convergence information calculation module 7 is used to match feature points in the measurement area and calculate section convergence information. In a deep, high-stress tunnel monitoring system in a metal mine, convergence information includes: convergence amount (maximum 42 mm), convergence rate (maximum 3.5 mm / day), convergence distribution (a bell-shaped distribution with maximum convergence in the middle of the roof and decreasing toward the sides), and convergence trend (exponential decay curve with rapid growth in the early stages and stabilization in the later stages).

[0119] Convergence calculation uses the difference method on the time series:

[0120] ,

[0121] ,

[0122] in: is the convergence amount at time t, in millimeters; is the cross-section point cloud data at time t; is the benchmark point cloud data in the initial state; is the convergence rate at time t, in mm / day; The system plots convergence contours based on the calculated results, visually demonstrating the deformation distribution. In this tunnel, the maximum convergence occurs in the middle of the roof (red area), corresponding to the location of roof delamination that occurs later, validating the system's predictive capabilities.

[0123] Section Convergence Trend and Status Assessment Module 8 is used to establish a section safety assessment model. In a deep soft rock tunnel application case in a coal mine, the model inputs include the current convergence information vector (containing five key parameters: maximum convergence, average convergence, maximum convergence rate, convergence acceleration, and heterogeneity index), the historical convergence information matrix for the past 30 days, and the environmental monitoring information vector (temperature, humidity, gas concentration, etc.).

[0124] Evaluation indicators are set as follows: an overall convergence rate warning value of 2.5 mm / day, a maximum convergence position convergence value of 50 mm, a non-uniformity index warning value of 0.6 (a higher value indicates greater non-uniformity), an acceleration coefficient warning value of 0.15 mm / day², and a minimum safety margin of 20%. If any indicator exceeds 80% of the warning value, the system enters a warning state.

[0125] In actual application, the system successfully predicted a local roof collapse by identifying a sudden increase in convergence acceleration (from 0.05 to 0.22 mm / day²), issuing a warning 48 hours in advance. The mine implemented reinforcement measures in a timely manner, avoiding the accident.

[0126] Cross-section display module 9 displays real-time monitoring data and cross-section convergence trend curves. In an application example at a large iron ore mine's surface monitoring center, this module displays: 3D visualization (using pseudo-color coding, with red indicating areas of significant deformation and blue indicating areas of minimal deformation); convergence trend curves for key measurement points (over the past 7 days and 30 days); graded alarm information (combining audio, video, and lighting); and decision-making recommendations (such as support parameter optimization suggestions).

[0127] The system utilizes a 55-inch 4K resolution display, combined with a touch-screen interface, allowing managers to intuitively understand the status of each tunnel section. In remote operation mode, the system supports viewing key data and alarm information on mobile devices via a secure VPN connection, enabling 24-hour uninterrupted monitoring.

[0128] Cross-sectional deformation feature description module 10 is used to describe deformation characteristics and construct deformation feature descriptors and association networks. In a monitoring system for a roadway above a stope in a deep metal mine, deformation feature descriptors include factors such as roof subsidence (8.5 mm), sidewall convergence (5.2 mm), floor uplift (2.8 mm), overall convergence direction (towards the stope), and convergence rate (1.2 mm / day).

[0129] The correlation network, comprised of 48 measurement points, identified 125 valid correlations: 86 direct, 24 indirect, 10 forced, and 5 sequential. Using this information, the system successfully identified a progressive deformation pattern caused by mining, moving from the side closest to the stope to the side further away. This provided valuable insights for support design.

[0130] The connection relationship between the modules of the system of the present invention is as follows: the laser projection module 2 for each measuring point, the circumferential measurement module 3 for the measuring point, the cross-section convergence information calculation module 7, the cross-section convergence trend and status assessment module 8, and the cross-section display module 9 are connected in sequence to form the main line of data processing; the measurement area information control module 1, the cross-section convergence information calculation module 7, the cross-section display module 9, and the laser projection module 2 for each measuring point are connected in sequence to form a control feedback loop.

[0131] The laser structured light projector and CMOS sensor group 5 are connected to multiple input ends of the cross-section convergence information calculation module 7, the cross-section convergence trend and status evaluation module 8, and the cross-section display module 9 to provide raw data; multiple input ends of the laser projection module 2 of each measuring point, the circumferential measurement module 3 of the measuring point, the measurement area contour alignment module 6, and the cross-section convergence information calculation module 7 are interconnected to achieve data sharing; the cross-section convergence information calculation module 7 is connected to the cross-section three-dimensional display module 9 to output the processing results; multiple output ends of the laser projection module 2 of each measuring point, the circumferential measurement module 3 of the measuring point, the deformation matching module 4, and the measurement area contour alignment module 6 are interconnected to form a complete data network.

[0132] In the application case of the transport tunnel of a fully mechanized mining face in a coal mine, the system deployed 6 measurement areas and a total of 48 measurement points, covering a horseshoe-shaped section of 5m×3.5m. The system operation process is as follows: the measurement area information control module 1 is first initialized, dividing the section into 6 sector-shaped measurement areas, each of which is evenly distributed with 8 measurement points; the laser projection module 2 generates a random spot coding projection signal at each measurement point, and the laser structured light projector and CMOS sensor group 5 perform synchronous projection and imaging every 5 minutes, with an acquisition resolution of 1280×1024 pixels; the measurement point circumferential measurement module 3 obtains feature information within a 20cm radius around each measurement point; the measurement point deformation matching module 4 extracts approximately 60 feature points for each area and matches them, with a matching success rate of 9 2%; the survey area contour registration module 6 establishes the constraint relationship between the survey areas, and the registration accuracy is better than 1mm; the section convergence information calculation module 7 processes the matching results and calculates the maximum section convergence of 7.5mm and the average convergence rate of 0.8mm / day; the section convergence trend and status assessment module 8 assesses the safety status, with a safety score of 73 points, which is at the attention level; the section deformation feature description module 10 identifies an overall convergence mode dominated by the roof; finally, the section display module 9 displays the monitoring results through a three-dimensional visualization interface, and recommends appropriately increasing the preload force of the roof anchor rods and strengthening the support at the corners of the tunnel.

[0133] Through the above system and method, the present invention realizes all-round, high-precision and intelligent monitoring of tunnel cross-section convergence, providing a strong guarantee for mine safety production.

[0134] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A tunnel cross-section convergence multi-view 3D reconstruction method, characterized by: include: A plane rectangular coordinate system is established on the tunnel cross section, and multiple measurement areas are arranged around the cross section. Each measurement area has a number of equal-height measurement points evenly distributed along the cross section elevation direction. At least one measurement point is shared by both adjacent measurement areas. A laser structured light projector is arranged at each measurement point. The normal direction of the laser structured light shadow surface is perpendicular to the tunnel contour line. A CMOS receiving camera is set next to each laser structured light projector. A threshold S2 is set for the overlapping portion of the shooting area of ​​each laser structured light projector and the adjacent CMOS camera. Multiple laser structured light groups are set around the cross section. Each laser structured light group includes a structured light projector and a CMOS receiving camera. The number of laser structured light groups is determined by the cross section size. Based on the laser beam of the projector, surface feature projection encoding is performed on each measuring point and a preset measuring area circumferentially of the measuring point. The CMOS receiving camera receives the projection echo and extracts feature points of the measuring point and the measuring area circumferentially thereof. The feature points of each measuring point and the measuring area circumferentially thereof are matched using an improved stereo matching algorithm to obtain depth feature point information of the measuring point in the observation area, establish cross-sectional point cloud data, and calculate cross-sectional convergence information. The cross-section convergence deformation is divided into local deformation and global deformation. The local deformation refers to the convergence deformation within the current measurement area, and the global deformation includes the convergence deformation transfer between the current measurement area and the adjacent measurement area. In the cross-section local deformation detection stage, each measurement point is used as a feature, and feature matching is performed within the observation area of ​​the cross-section laser structured light group and the CMOS camera group to obtain the cross-section convergence information of each measurement area. In the cross-section global deformation detection stage, the cross-section convergence information of the current measurement area is obtained, and at the same time, the deformation features of the adjacent measurement points are described according to the threshold S2. The deformation association between the measurement points is established based on the deformation feature description. After forming local deformation compensation, the point cloud registration method is used to perform cross-section matching to obtain the cross-section global convergence information. Based on current convergence information, historical convergence information, and real-time environmental monitoring information, a section safety evaluation model is established to obtain a section deformation assessment, issue a warning for abnormal section deformation, and display real-time monitoring data and section convergence trend curves, and provide reinforcement recommendations for subsequent section deformation.

2. The method according to claim 1, wherein: The measurement area preset in the circumferential direction of the measurement point is within a range with the measurement point as the center and R as the radius.

3. The method according to claim 2, wherein: The measuring points are arranged in sequence along the elevation direction of the cross section.

4. The method according to claim 1, wherein: The fact that at least one measuring point between two adjacent measuring areas is shared by both areas means that at least one pair of adjacent edges between the measuring areas overlap, the overlapping length of the adjacent edges is not less than R, and the overlapping area of ​​the measuring points is within the overall cross-sectional contour.

5. The method according to claim 1, wherein: The improved stereo matching algorithm establishes a mathematical model to achieve accurate matching of feature points and obtain depth feature point information by combining parameters such as disparity value, laser beam cross-sectional area, camera intrinsic parameters, camera focal length, camera spacing, lens focal length, and the distance between the camera reference point and the laser structure light source point.

6. The method according to claim 1, wherein: During the overall deformation detection, the cross-sectional contour data of adjacent measurement areas are divided according to the deformation characteristics of different measurement points into contour data of different measurement points. The deformation data of the measurement points in the current measurement area are matched with the deformation data of the measurement points in the adjacent measurement areas to complete the deformation point matching from the current measurement point to the adjacent position measurement area. Through the matching of the measurement point deformation data, the constraint relationship between the cross-sectional deformation points is established, the overall deformation of the cross section is established, and the overall convergence information of the cross section is obtained.

7. The method according to claim 1, wherein: There are two types of preset measurement areas at the measurement points: one is the circumferential measurement area of ​​the measurement point on the cross section; the other is the measurement area between measurement areas and between measurement points on the cross section. The two measurement areas partially overlap.

8. The method according to claim 1, wherein: When the section state and deformation information is output, the measurement points of the section, the distribution density of the measurement points, the average convergence of the measurement points, the maximum convergence of the measurement points, the convergence change ratio of the measurement points, the convergence trend of the measurement points, the overall convergence of the section, the convergence curve, the convergence distribution diagram, the safety status of the measurement points, and the safety status of the section state are used as parameters to display the real-time monitoring data and the section convergence trend curve, give reinforcement suggestions for subsequent section deformation, establish a section safety evaluation model, and display the safety hazards of the section in a graded manner based on the established section safety evaluation model.

9. The method according to claim 1, wherein: The deformation feature description includes deformation amount, deformation direction, deformation rate and deformation distribution. A deformation association network between measurement points is constructed through the deformation feature description; the deformation association network includes direct association, indirect association, forced association and temporal association, which is used to analyze the spatial transmission law of deformation and assist in deformation trend prediction.

10. A tunnel cross-section convergence multi-view 3D reconstruction monitoring system using the method according to any one of claims 1 to 9, characterized in that: The system comprises a measurement area information control module, laser projection modules for each measurement point, a circumferential measurement module for the measurement point, a deformation matching module for the measurement point, a laser structured light projector and a CMOS sensor group, a measurement area contour registration module, a cross-section convergence information calculation module, a cross-section convergence trend and state evaluation module, a cross-section display module, and a cross-section deformation feature description module. The laser projection modules for each measurement point, the circumferential measurement module for the measurement point, the cross-section convergence information calculation module, the cross-section convergence trend and state evaluation module, and the cross-section display module are connected in sequence. The measurement area information control module, the cross-section convergence information calculation module, the cross-section display module, and the laser projection modules for each measurement point are connected in sequence. The laser structured light projector and the CMOS sensor group are connected to multiple input ends of the cross-section convergence information calculation module, the cross-section convergence trend and state evaluation module, and the cross-section display module. The laser projection modules for each measurement point, the circumferential measurement module for the measurement point, the measurement area contour registration module, and the cross-section convergence information calculation module are connected to multiple input ends. The cross-section convergence information calculation module is connected to The cross-sectional three-dimensional display module includes multiple output terminals of the laser projection modules for each measurement point, the circumferential measurement module for each measurement point, the deformation matching module for each measurement point, and the contour matching module for the measurement area. The measurement area information control module is configured to obtain circumferential information of the cross section, overlap information between adjacent measurement areas, collect constraint relationships between measurement areas, and control measurement area information. The laser projection modules for each measurement point are configured to generate coded projection signals to implement projection coding of surface features of the measurement area. The circumferential measurement module for each measurement point on the cross section and a measurement area preset in the circumference of the measurement point is configured to collect feature information of the measurement point. The deformation matching module for each measurement point is configured to establish feature information for each measurement point and implement feature matching of each measurement point on the cross section. The laser structured light projector and CMOS sensor group are configured to collect cross-sectional contours of each measurement area and measurement results of the measurement points. The contour matching module for the measurement area is configured to establish constraint relationships between measurement points on the cross section. The cross-sectional convergence information calculation module is configured to match feature points in the measurement area and calculate cross-sectional convergence information. The section convergence trend and status assessment module is used to establish a section safety assessment model; The cross-section display module is used to display real-time monitoring data and cross-section convergence trend curves.

Images