A coal mine tunnel segmented continuous mapping method and system supporting long-period operation

Abstract

The application provides a kind of coal mine roadway segmented continuous mapping method and system supporting long-period operation, belong to coal mine intelligent mining and roadway driving surveying and mapping technical field.The application realizes high-precision trajectory estimation and roadway three-dimensional modeling by laser radar, IMU and odometry three sensors parallel data acquisition, specifically: with existing tight coupling LIO estimation as pose solving base, introduce key frame-natural day section double-layer management, odometry residual constraint, section-level optimization, global optimization and snapshot-level power failure recovery, realize trajectory continuity maintenance in long-period driving process, drift error real-time detection and gradual correction, continuous generation and update of local and global map in roadway, support segmented management in long-period construction process, history backtracking and stable operation.The application is suitable for mine unmanned driving environment perception, coal mine driving real-time surveying and mapping and other fields, and can realize continuous mapping, segmented management and trajectory maintenance in GNSS signal-free environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent coal mining and tunnel excavation mapping technology, specifically a method and system for segmented continuous mapping of coal mine tunnels that supports long-term operation. Background Technology

[0002] Currently, manual cross-section mapping, total station fixed-point measurement, or single-point laser scanning equipment are commonly used for roadway measurement and 3D mapping in coal mines and tunnel excavation faces. However, these methods have the following shortcomings: 1. Low efficiency of manual mapping: It requires personnel to enter the mine, posing safety risks, and efficiency and accuracy are affected by human factors. 2. Total station / single-point laser scanning: It often requires multiple standing and stitching, resulting in low efficiency and difficulty in achieving continuous real-time mapping, especially during long-distance, long-cycle tunneling, where data connection is difficult and trajectory continuity cannot be guaranteed. 3. Mobile laser scanning scheme: Although there are existing SLAM-based point cloud modeling methods, in environments without GNSS signals and with heavy dust interference, there are still problems such as large positioning errors, significant cumulative drift, and serious trajectory deviation, making it difficult to maintain the consistency of long-cycle roadway mapping. 4. Existing laser-IMU schemes: such as LIO-SAM, Fast-LIO, Faster-LIO, etc., although they can achieve tightly coupled positioning mapping, lack specialized optimization designs for long-cycle operation and segmented continuous mapping in underground tunneling. During long-term operation, the system needs to cope with anomalies such as power outages, restarts, and sensor failures. The accumulated error is difficult to control effectively, and there is a lack of stage management and odometer correction strategies. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for segmented continuous mapping of coal mine roadways that supports long-term operation, so as to solve the problem that the existing technology cannot meet the requirements of real-time, safe, continuous and manageable long-term segmented continuous mapping in underground coal mine roadway excavation.

[0004] This invention is implemented as follows: A segmented continuous mapping system for coal mine roadways that supports long-term operation includes: The lidar, connected to the data processing unit, is used to collect three-dimensional point cloud data of the coal mine roadway environment and send it to the data processing unit. The IMU, connected to the data processing unit, is used to collect acceleration and angular velocity data of the tunneling machine and send the collected data to the data processing unit. The wheel odometer, connected to the data processing unit, is used to collect the wheel speed of the tunneling machine and send the collected data to the data processing unit. The data processing unit is connected to the lidar, IMU, and wheeled odometer respectively. It processes the data sent by the lidar and IMU, performs tightly coupled LIO estimation, and triggers keyframe generation based on displacement thresholds. It divides the data into segments according to natural days or construction shifts. The first and last keyframes within a segment store complete data, while the middle keyframes are downsampled data. Pose graph constraints are established between adjacent segments through the first and last keyframes. The data processing unit also supplements the local optimization process with odometer residual constraints based on the data sent by the wheeled odometer. Storage units are used to store keyframes and stage data; and The power supply unit is used to provide the required operating voltage to the above-mentioned units.

[0005] Preferably, the keyframe includes point cloud data, IMU fused pose data, keyframe number, timestamp, and cumulative mileage information; the segment includes the current keyframe set, raw sensor data, snapshot index, and start and end pose.

[0006] Preferably, the displacement threshold is 0.5 m.

[0007] Preferably, the data processing unit uses Fast-LIO2 as the tightly coupled LIO state estimation base.

[0008] This invention also provides a method for segmented continuous mapping of coal mine roadways that supports long-term operation, comprising the following steps: (1) Multi-source data acquisition: The lidar acquires three-dimensional point cloud data of the coal mine roadway environment, the IMU acquires the acceleration and angular velocity data of the tunneling machine, and the wheel odometer acquires the wheel speed of the tunneling machine; the data acquired by the lidar, IMU and wheel odometer are all sent to the data processing unit. (2) The data processing unit performs tightly coupled LIO state estimation based on the data sent by the lidar and IMU, and calculates the cumulative LIO displacement; (3) The data processing unit determines whether the cumulative displacement of LIO is greater than the displacement threshold; if not, it continues to accumulate displacement; if so, it triggers the generation of keyframes. (4) The data processing unit performs odometer residual constraints based on the data sent by the wheel odometer in order to optimize the local pose graph; (5) The data processing unit divides the construction data into segments according to natural days or construction shifts. Each segment contains all keyframes, raw sensor data, start and end poses, and snapshot indexes for the day or current construction shift. (6) The data processing unit performs inter-segment ICP registration, pose graph optimization, historical backtracking and power failure recovery based on segment data.

[0009] Preferably, step (4) specifically involves: (4-1) The data processing unit obtains the cumulative displacement of the odometer based on the data sent by the wheel odometer. L odom , i ; (4-2) The data processing unit calculates the cumulative displacement of LIO. L LIO , i Cumulative displacement with odometer L odom , i The difference Δ between L i = L LIO , i - L odom , i ; (4-3) The data processing unit injects the residuals into the objective function of the sliding window optimizer: ; where e ICP The point-to-plane registration residual between the keyframe and the local map, e IMU The IMU pre-integration residual from adjacent keyframes, Δ L i This represents the drift residual of the current keyframe over the cumulative travel distance.

[0010] Preferably, λ3 is Δ L i The weights, λ3, are determined by the baseline weights λ. 3,0 With adaptive factor w odom Multiplying them together, we get λ3 = λ 3,0 ·w odom , where w odom The calculation formula is: exp (-| ΔL i - μ r | / σ r ); where Δ L i Let μ be the odometry drift residual of the i-th keyframe. r σ is the historical residual mean. r The standard deviation is denoted as .

[0011] Preferably, the sliding window in step (4-3) covers the most recent 10 to 50 keyframes.

[0012] Preferably, in step (3), a snapshot is saved synchronously for each keyframe generated; the snapshot includes the current keyframe sequence pose, current day segment metadata, nearest point cloud cache, odometer cumulative reading and optimizer status; at the same time, scan data that has not yet formed a complete keyframe is written to a temporary cache.

[0013] Preferably, when restarting after a power outage, the data processing unit first searches for the most recent valid snapshot, prioritizes loading the latest version, and if the verification fails, it rolls back to the backup. Then, it reads the odometer increments before and after the power outage, adds the temporary cache to the end of the trajectory, and continues mapping.

[0014] The present invention has the following beneficial effects: 1) This invention proposes a cumulative displacement residual constraint based on wheeled odometers as a supplementary observation for LIO estimation; it designs an adaptive weight adjustment strategy to dynamically adjust the odometer confidence level according to slippage detection and ground conditions; it effectively suppresses lateral drift in long straight tunnels, reducing lateral drift by more than 60% compared to the pure LIO scheme.

[0015] 2) This invention proposes a two-level data organization method: keyframe level (0.5m interval) and daily segment level (by calendar day or construction shift). The keyframe layer implements real-time sliding window optimization, while the daily segment level implements offline global optimization; it supports power outage snapshot recovery, and the system can be restored to the previous state within seconds after a restart.

[0016] 3) The first and last keyframes of each daily segment are used as pose graph nodes, and the relative poses between segments are used as edges; incremental Levenberg-Marquardt optimization is adopted, making local adjustments only to newly added segments and their neighborhoods. The computational complexity of global optimization is reduced from O(N^2) to O(N^2). 2 The value is reduced to O(kN), where k is the neighborhood window size (usually ≤5).

[0017] 4) This invention designs a three-source synchronization mechanism based on timestamp interpolation of laser-IMU-odometer and proposes an odometer slippage detection algorithm: by comparing LIO speed estimation with wheel speed, anomalies are identified in real time; after anomaly detection, the corresponding observations are automatically downweighted or removed to ensure fusion robustness.

[0018] The technical solution of this invention is applicable to fields such as environmental perception for unmanned driving in mines and real-time mapping of coal mine tunneling. Attached Figure Description

[0019] Figure 1 This is the system architecture diagram of the present invention.

[0020] Figure 2 This is a schematic diagram of the data processing of the present invention.

[0021] Figure 3 This is the flowchart of the core algorithm of this invention.

[0022] Figure 4 This is a flowchart of the keyframe-natural day segment generation process of the present invention.

[0023] Figure 5 This is a diagram of the dual-layer management architecture of keyframes and natural day segments in this invention.

[0024] Figure 6 This is a schematic diagram of incremental optimization of inter-segment pose graphs according to the present invention.

[0025] Figure 7 This is a flowchart of the odometer calibration process of the present invention.

[0026] Figure 8 This is a diagram of the adaptive fusion mechanism for odometer residuals in this invention. Detailed Implementation

[0027] The coal mine roadway segmented continuous mapping system provided by this invention, which supports long-term operation, mainly includes equipment such as a base station, lidar, inertial measurement unit (IMU), wheeled odometer, data processing unit, power supply unit, and storage unit.

[0028] The main function of the base station is for network switching and data forwarding, forming a communication bridge between the tunneling machine's internal and external control centers. The front of the tunneling machine houses the drum cutting device, while the middle and rear sections integrate the control cabin and track drive mechanism. Based on this, sensor supports and protective devices are added to mount lidar, IMU, and auxiliary equipment, thereby enabling tunnel modeling and positioning during the tunneling process. The overall layout maintains the original tunneling performance while also considering the sensor's observation field of view and protection requirements.

[0029] The LiDAR selected is the Mid-360, a wide-angle, scattered LiDAR with a 360° horizontal and 60° vertical field of view. Mounted on the roof in front of the tunneling machine, it collects 3D point cloud data of the tunnel environment and scans the tunnel cross-section in real time. The LiDAR's position is optimized to ensure a large horizontal and vertical field of view, covering the front and sidewall areas, providing complete data for point cloud stitching and tunnel modeling. The IMU uses an HL14R2 high-frequency 9-axis inertial measurement unit, which collects acceleration and angular velocity data at high frequency to achieve real-time attitude calculation, motion prediction, and distortion correction. The IMU is designed with an adaptive sampling rate, reducing its sampling frequency when stationary and increasing it to 200Hz when in motion. A wheeled odometer is used to accumulate the tunneling machine's displacement. It measures wheel speed through encoders or photoelectric sensors to calculate the cumulative displacement, which serves as a long baseline constraint for global correction. The wheeled odometer data undergoes Kalman filtering preprocessing to remove high-frequency noise and transient jitter. The three sensors—LiDAR, IMU, and wheeled odometer—are synchronized using hardware timestamps, achieving millisecond-level time alignment accuracy.

[0030] Combination Figures 1-3 Data collected by the LiDAR, IMU, and wheeled odometer are all transmitted to the data processing unit. The data processing unit includes an embedded Linux board. It performs noise reduction, dust removal, and invalid point removal on the point cloud data collected by the LiDAR; performs adaptive downsampling based on local geometric features, with voxel size adjustable from 0.05 m to 0.2 m to balance accuracy, real-time performance, and storage efficiency; uses IMU data to correct point cloud motion distortion, ensuring the accuracy of the point cloud set; the LiDAR and IMU acquire data synchronously, and hardware timestamps ensure the accuracy of data projection and distortion correction; the data processing unit is also used for odometer residual constraint optimization, natural day segment management, and global map stitching. The power supply unit supplies power to the LiDAR, IMU, Linux board, and other accessories. The storage unit stores keyframes and stage data, supporting power-off continuation of computation.

[0031] This invention preferably uses Fast-LIO2 as the existing tightly coupled state estimation base, and superimposes keyframe-natural day segment management, odometer residual constraints, inter-segment pose map optimization and snapshot recovery mechanism on this base to achieve: maintaining trajectory continuity during long-cycle tunneling, real-time detection and progressive correction of drift errors, continuous generation and updating of local and global tunnel maps, and support for segmented management, historical backtracking and stable operation during long-cycle construction.

[0032] Specifically, the system first uses the IMU to perform forward propagation between adjacent lidar scans to obtain the initial pose of the current scan; then it establishes a residual between the current point cloud after distortion correction and the planar / edge features in the local map, solves the pose increment through iterative error state update, and outputs the pose, position, velocity and IMU bias estimate of the current key frame.

[0033] After keyframes are generated, the tightly coupled estimation results are used as the initial relative poses of adjacent keyframes. Point-to-plane ICP registration is performed on the downsampled point clouds of the two frames: first, the nearest neighbor correspondence is established, then outliers with excessive distance or inconsistent normals are removed, and then the rigid body transformation is solved iteratively to minimize the registration error until the pose increment is lower than the threshold or the maximum number of iterations is reached.

[0034] Combination Figure 4The keyframe and natural day segment modeling method is as follows: The system is started, the natural day segment is initialized, and the data processing unit receives data collected by the lidar and IMU, calculates the cumulative displacement of the tunneling machine, and triggers keyframes based on a displacement threshold (e.g., generating one frame every 0.5 m). The data processing unit determines whether the cumulative displacement exceeds the 0.5 m threshold. If not, it continues to accumulate displacement; if so, it generates a keyframe. In other words, the system automatically saves a keyframe whenever the tunneling machine advances 0.5 m along the tunnel. A keyframe includes at least the current point cloud data, IMU fused pose data, keyframe number, timestamp, and cumulative mileage information. During point cloud acquisition, the system triggers keyframes based on the displacement threshold (e.g., generating one frame every 0.5 m) to ensure sufficient overlap between any two keyframes without redundant calculations due to excessively small intervals.

[0035] Combination Figure 5 The data is divided into segments according to natural days (or construction shifts). When saving each segment, the first and last keyframes save complete data, the middle keyframes save downsampled data, and the pose graph constraints between adjacent segments are established through the first and last keyframes (obtained through ICP registration).

[0036] When the date changes or the preset segmentation conditions are met, the system encapsulates the current keyframe set, raw sensor data, snapshot index, and start and end poses into a daily segment, enabling inter-segment ICP and pose graph optimization to run directly on a daily segment basis. Keyframes are responsible for "accurate calculations," while natural daily segments are responsible for "management, splicing, and retrieval after power failure." This two-layer organization method is specifically designed for long-cycle downhole mapping scenarios.

[0037] Combination Figure 6 When a new day segment is added, only that stage and its preceding 4 segments (a total of 5 segments, i.e., neighborhood window k=5) are optimized, using the pose T of the first and last keyframes of each day segment. i As a pose graph node, the relative pose ΔT of adjacent segments obtained by inter-segment ICP registration is used. ij As edges in the pose graph, the optimization objective is to minimize the inter-segment registration error. Where ⊕ represents pose composition, This represents the inverse operation of pose; the Levenberg-Marquardt algorithm is used iteratively to solve for the pose increment; after convergence, the keyframe pose of the affected segment is updated. A traditional method is used for global optimization: the complexity is O(N). 2 N is the total number of segments. Using the method of this invention for global optimization, the complexity is O(kN), where k is the neighborhood window size (k=5). When N=100, the computational load is reduced by 95%. It achieves a balance between "local optimization and global consistency"; supports online incremental optimization without offline batch processing; and reduces the global closed-loop error from the meter level to the decimeter level.

[0038] The system encapsulates construction data into a segment (named as SEG_YYYYMMDD) by natural day or construction shift. The segment contains all keyframes of the day or current shift, raw IMU data, odometer data, start and end poses, local map summary and snapshot index. This "day segment" is not only used for archiving, but is directly embedded into the optimization, recovery and storage process, making it a unified processing boundary for inter-segment ICP registration, pose map optimization, historical backtracking and power outage recovery.

[0039] Keyframe partitioning ensures local point cloud registration accuracy, while daily segment partitioning not only facilitates data management but, more importantly, provides fixed technical boundaries for inter-segment registration, pose graph node definition, snapshot archiving, and power outage recovery. The combination of these two approaches allows the system to simultaneously achieve both "short-scale geometric accuracy" and "long-term manageability."

[0040] Moreover, based on the key frame-day segment dual-layer management, the system memory usage is reduced from linear growth O(N) to constant O(1), supporting infinitely long cycle operation; the power outage recovery time is reduced from minutes to seconds, improving project availability; segmented management facilitates quality traceability and data archiving according to construction shifts.

[0041] During underground tunnel excavation in coal mines, the accumulated pose data based on IMU and lidar may experience some degree of drift due to the monotonous environmental characteristics and slow excavation speed. To ensure the global consistency and continuity of the keyframe trajectory, this invention injects wheel odometer data from the tunneling machine as an auxiliary constraint into the keyframe optimization process. Although wheel odometer data may have accumulated errors, it can provide reliable incremental travel information over short distances.

[0042] Combination Figure 7 and Figure 8 Based on LIO data, the system automatically generates a keyframe every time the tunneling machine advances 0.5 m along the roadway. For each keyframe, the system reads the accumulated displacement L from the odometer. odom Wheel-mounted odometers provide a reference for the tunneling machine's travel distance and are used for segment-level trajectory constraints.

[0043] For the i-th keyframe, calculate the LIO cumulative displacement. L LIO , i Cumulative displacement with odometer L odom , i The difference between them; Δ L i = L LIO , i - L odom ,i , where Δ L i It represents the drift trend of the current keyframe over the cumulative travel distance.

[0044] Injecting residuals into the objective function of the sliding window optimizer: .

[0045] Where e ICP The point-to-plane registration residual between the keyframe and the local map, e IMU The IMU pre-integration residual from adjacent keyframes, Δ L i The drift residual of the current keyframe over the cumulative travel distance; the three types of residuals are first normalized according to their respective dimensions, and then outliers are suppressed by Huber kernel or threshold mechanism.

[0046] The residual weighting can be dynamically adjusted according to the sensor's reliability. For example, under normal operating conditions, λ1=1.0, λ2 is adaptively adjusted according to the IMU noise covariance, and λ3=0.8. When |Δ L i When the drift gradually and continuously increases and exceeds the drift alarm threshold, and the odometer does not exhibit slippage or other abnormalities, λ3 should be appropriately increased to enhance the suppression of long-range drift. However, when the odometer residual deviates from its historical distribution due to sudden changes such as slippage or idling, the following adaptive weighting function should be used. w odom Automatically reduce its weight. These two correspond to the two scenarios of "long-term gradual drift requiring stronger constraints" and "instantaneous anomalies requiring reduced trust," respectively.

[0047] Design an adaptive weighting function: w odom = exp (-| ΔL i - μ r | / σ r ) Where μ r σ is the historical residual mean. r The standard deviation is used. The weights are automatically reduced when the residuals deviate from the historical distribution.

[0048] Compared to the fixed-weight scheme, adaptive weighting reduces positioning error by 40% in slippage scenarios; compared to completely eliminating odometers, it reduces lateral drift by 60% in long, straight tunnels. It achieves robust fusion that "enhancees when available and degrades when abnormal."

[0049] Combination Figure 5The system sets up a sliding window for local optimization. The sliding window typically covers the most recent 10 to 50 key frames, and in a specific embodiment, 30 frames can be used. The window simultaneously retains point cloud registration constraints, IMU pre-integration constraints, and odometry residual constraints. The local map is continuously updated as the window scrolls, and the local drift within the segment is controlled within 2 cm.

[0050] The optimized output includes corrected keyframe poses, window trajectories, and segment maps. If a day segment ends, segment-level / global pose map optimization is performed again using the start and end poses of the day segment as nodes.

[0051] From an engineering implementation perspective, odometry constraints are not always forcibly enabled. Instead, they participate in optimization only when odometry timestamps are continuous, encoder increments are stable, and no significant slippage is detected. If a communication interruption, count jump, or machine idle is detected, the system degenerates into a local optimization mode that relies solely on point cloud registration and IMU pre-integration (i.e., the odometry residual term is no longer injected into the objective function, equivalent to its weight λ3 being set to 0), thus ensuring that the trajectory is not interrupted rather than forcibly correcting errors. It should be noted that the "hard rejection" (weights reset to 0) here is different from... Figure 8 The adaptive weighting function in the middle class has two different mechanisms for the "continuous deweighting" of the odometer (when the residual is abnormal but the odometer data is still valid, the lower limit of the adaptive weighting factor is about 0.2): the former is for hard faults that cause the odometer data to fail (communication interruption, count jump, machine idling), in which case the data is unreliable and is completely removed; the latter is for soft anomalies that are valid but the residual is large (such as slight slippage), in which case certain constraints are still retained to avoid information waste.

[0052] The optimized keyframe poses, point clouds, and segment snapshots are saved. Segment snapshots are used for power outage recovery, historical backtracking, and local / global optimization. During algorithm execution, a snapshot is saved synchronously for each keyframe generated; the snapshot includes at least the current keyframe sequence pose, current day segment metadata, the most recent point cloud cache, odometry cumulative readings, and optimizer status. Simultaneously, scan data that has not yet formed a complete keyframe is written to a temporary cache.

[0053] Upon restarting after a power outage, the system first searches for the most recent valid snapshot, prioritizing the loading of the latest version. If verification fails, it reverts to the backup. Subsequently, it reads the odometer increments before and after the power outage, appends the temporary cache to the end of the track, and continues mapping without requiring complete re-initialization. This recovery method differs from conventional methods like "reopening the project file" or simple automatic saving. This invention requires snapshot granularity to be consistent with keyframes, employs a double buffer (latest / backup) to prevent partial write corruption during power outages, and combines the temporary cache and odometer increments to reconnect the track near the breakpoint during recovery.

[0054] The system collects data in parallel using three independent sensors: lidar, IMU, and odometer, to achieve high-precision trajectory estimation and 3D modeling of tunnels.

[0055] Specifically, the system maintains the current state x. k = {R k , p k , v k , b g , k , b a , k}, where R, p, and v represent attitude, position, and velocity, respectively, and b g b a These represent the gyroscope and accelerometer biases, respectively. For each frame of LiDAR data received, forward prediction is first performed based on continuous integration of the IMU to obtain the initial state of the current scan; then, iterative correction is performed by combining the distortion-free point cloud and the ICP / feature residuals of the local map, outputting the attitude, position, velocity, bias, and cumulative odometer estimate of the current keyframe.

[0056] For example, when the system outputs a keyframe KEY i At the same time, the day segment number, three-dimensional pose, cumulative mileage, local map index and snapshot number of the keyframe can be obtained. These results are used for visualization, inter-segment stitching, power outage recovery and historical backtracking, respectively.

[0057] The daily construction data is uniformly classified into a natural day segment (SEG_YYYYMMDD). Keyframes and metadata within the segment are archived to support power outage recovery. Local maps are generated within the segment, and the segments are optimized and stitched together into a global map of the entire tunnel through ICP and pose graph. The final output is a visualization model, point cloud file, or CAD file.

[0058] To achieve a clear and intuitive display of continuous mapping results, this invention employs a combination of Qt's Quick module and VTK for full-segment rendering. The system uses the VTK rendering pipeline to display the keyframe point clouds, segment maps, and overall roadway global map generated from the mapping process in 3D, while Qt provides the Quick interface framework to enable interactive operation and visual control.

[0059] Specific functions include: Modular design: The rendering module is decoupled from the data processing, keyframe management, and natural day segment management modules. It receives point cloud, trajectory, and model data through interfaces to achieve front-end and back-end separation.

[0060] Real-time data flow: The front end receives keyframes and incremental updates of the daily map, and continuously renders the local map and trajectory in real time through the VTK pipeline to ensure real-time visualization of the tunneling process.

[0061] Interactive controllability: Through the interface space provided by Qt, the user can rotate, zoom and slice the view, allowing users to flexibly observe the three-dimensional structure and trajectory information of the tunnel.

[0062] Scalability: The design supports loading data from multiple days or selecting different keyframes for display, real-time historical backtracking and local reconstruction, and also supports exporting point clouds or CAD files.

[0063] The keypoint cloud or natural day segment data generated by the backend modeling system is received and transmitted to the frontend. Subsequently, the data is converted into an internal data structure supported by VTK and the rendering pipeline is updated incrementally to avoid the performance overhead of global redraws. Then, the rendering engine renders the updated data and refreshes the 3D display window in real time. Based on this, users can interact with the interface, such as rotating, scaling, sectioning, or switching keyframes. The system dynamically adjusts rendering parameters based on user input, achieving an interactive closed loop for data visualization.

[0064] The front end adopts a layered modular design, decoupling the information display part (QML) from the 3D rendering (VTK) and connecting with the back end (LIOBackend) through the data management layer (DataManager / KeyFrameManager / SegmentManager).

[0065] This invention achieves visualized management and optimized human-computer interaction for coal mine tunneling processes through interface design. Specific effects include: 1) Centralized information display: Integrating tunneling progress, keyframe / day segment information, environmental monitoring data, and operational status into a unified interface, reducing the need for operators to switch between multiple systems. 2) Real-time early warning and decision support: The interface highlights warnings when gas concentration exceeds limits, equipment malfunctions, or modeling errors are too large, allowing operators to immediately obtain critical information and take timely measures to effectively reduce safety risks. 3) Multi-dimensional progress visualization: Tunneling progress, energy consumption changes, and modeling errors are intuitively displayed through progress bars, graphs, and statistical charts, making the relevant data easy to understand and facilitating rapid decision-making. 4) Integration of historical and real-time data: The interface supports historical queries and comparisons of keyframe and day segment data, enabling the correlation between real-time modeling results and historical data for easy backtracking and quality assessment. 5) Convenient operation and scalability: Adopting a modular layout and unified entry point, users can quickly switch between different monitoring modules or export mapping data, improving operational efficiency.

[0066] Global map generation: The local maps within each day's segment are stitched together to form a global map of the entire tunnel, ensuring continuity between segments and global consistency; inter-segment registration and pose optimization are performed on the local maps of each natural day's segment, supporting real-time updates or offline end-point optimization; a visualization model, global point cloud file, or CAD file is generated to facilitate further analysis and construction management.

[0067] Data management: meta_data.json, keyframe point clouds and segment map files are saved for each daily segment; power-off resume calculation is supported, the latest segment snapshot and keyframe state are read to continue mapping, and even if the wheel odometer input is temporarily unavailable, the local trajectory can still be saved continuously; daily segment data can be used for historical archiving, construction analysis and global optimization.

[0068] Experimental verification: the system operated continuously for 7 consecutive natural days on a fully mechanized excavation face, with a cumulative tunneling distance of approximately 84 m, generated 168 keyframes and encapsulated 7 natural day segments. Existing measurement results show that: the maximum local drift within a segment is approximately 1.8 cm, the global trajectory error is approximately 7.3 cm, the average recovery time after power-off restart is approximately 4.2 s, and the total data volume is approximately 42 GB.

[0069] The comparison of technical effects between the method of the present invention and the traditional LIO scheme is shown in Table 1 below.

[0070] Table 1 Comparison table of technical effects between the method of the present invention and the traditional LIO scheme

[0071] It can be seen from the above that the present invention has the following advantages: 1) Long-period stability: supports continuous operation for several months, with constant memory occupation and no performance degradation; 2) Drift control capability: through odometer constraint + inter-segment optimization, the 1 km cumulative drift is reduced by more than 70% (see Table 1); 3) Engineering availability: supports fast recovery after power-off, segmented management and abnormal self-adaptation, meeting the complex underground working conditions; 4) Computing efficiency: adopts an incremental optimization strategy, and the global optimization time is reduced by 95%.

Claims

1. A segmented continuous mapping system for coal mine roadways that supports long-term operation, characterized in that, include: The lidar, connected to the data processing unit, is used to collect three-dimensional point cloud data of the coal mine roadway environment and send it to the data processing unit. The IMU, connected to the data processing unit, is used to collect acceleration and angular velocity data of the tunneling machine and send the collected data to the data processing unit. The wheel-type odometer, connected to the data processing unit, is used to collect the wheel speed of the tunneling machine and send the collected data to the data processing unit. The data processing unit is connected to the lidar, IMU, and wheeled odometer respectively. It processes the data sent by the lidar and IMU, performs tightly coupled LIO estimation, and triggers keyframe generation based on displacement thresholds. It divides the data into segments according to natural days or construction shifts. The first and last keyframes within a segment store complete data, while the middle keyframes are downsampled data. Pose graph constraints are established between adjacent segments through the first and last keyframes. The data processing unit also supplements the local optimization process with odometer residual constraints based on the data sent by the wheeled odometer. Storage units are used to store keyframes and stage data; and The power supply unit is used to provide the required operating voltage to the above-mentioned units.

2. The segmented continuous mapping system for coal mine roadways supporting long-cycle operation as described in claim 1, characterized in that, The keyframe includes point cloud data, IMU fused pose data, keyframe number, timestamp, and cumulative mileage information; the segment contains the current keyframe set, raw sensor data, snapshot index, and start and end poses.

3. The segmented continuous mapping system for coal mine roadways supporting long-cycle operation as described in claim 1, characterized in that, The displacement threshold is 0.5 m.

4. The segmented continuous mapping system for coal mine roadways supporting long-cycle operation as described in claim 1, characterized in that, The data processing unit uses Fast-LIO2 as the tightly coupled LIO state estimation base.

5. A method for segmented continuous mapping of coal mine roadways that supports long-term operation, characterized in that, Includes the following steps: (1) Multi-source data acquisition: The lidar acquires three-dimensional point cloud data of the coal mine roadway environment, the IMU acquires the acceleration and angular velocity data of the tunneling machine, and the wheel odometer acquires the wheel speed of the tunneling machine; the data acquired by the lidar, IMU and wheel odometer are all sent to the data processing unit. (2) The data processing unit performs tightly coupled LIO state estimation based on the data sent by the lidar and IMU, and calculates the cumulative LIO displacement; (3) The data processing unit determines whether the cumulative displacement of LIO is greater than the displacement threshold; if not, it continues to accumulate displacement. If so, keyframe generation is triggered; (4) The data processing unit performs odometer residual constraints based on the data sent by the wheel odometer in order to optimize the local pose graph; (5) The data processing unit divides the construction data into segments according to natural days or construction shifts. Each segment contains all keyframes, raw sensor data, start and end poses and snapshot indexes for the day or current construction shift. (6) The data processing unit performs inter-segment ICP registration, pose graph optimization, historical backtracking and power failure recovery based on segment data.

6. The method for segmented continuous mapping of coal mine roadways supporting long-cycle operation according to claim 5, characterized in that, step( 4) Specifically: (4-1) The data processing unit obtains the cumulative displacement of the odometer based on the data sent by the wheel odometer. L odom , i ; (4-2) The data processing unit calculates the cumulative displacement of LIO. L LIO , i Cumulative displacement with odometer L odom , i The difference Δ between L i = L LIO , i - L odom , i ; (4-3) The data processing unit injects the residuals into the objective function of the sliding window optimizer: ; where e ICP The point-to-plane registration residual between the keyframe and the local map, e IMU The IMU pre-integration residual from adjacent keyframes, Δ L i This represents the drift residual of the current keyframe over the cumulative travel distance.

7. The method for segmented continuous mapping of coal mine roadways supporting long-cycle operation according to claim 6, characterized in that, λ3 is Δ L i The weights, λ3, are determined by the baseline weights λ. 3,0 With adaptive factor w odom Multiplying them together, we get λ3 = λ 3,0 ·w odom , where w odom The calculation formula is: exp (-| ΔL i - μ r | / σ r );in, ΔL i Let μ be the odometry drift residual of the i-th keyframe. r σ is the historical residual mean. r The standard deviation is denoted as .

8. The method for segmented continuous mapping of coal mine roadways supporting long-cycle operation according to claim 6, characterized in that, The sliding window in step (4-3) covers the most recent 10 to 50 keyframes.

9. The method for segmented continuous mapping of coal mine roadways supporting long-cycle operation according to claim 5, characterized in that, In step (3), a snapshot is saved synchronously for each keyframe generated; the snapshot includes the current keyframe sequence pose, current day segment metadata, nearest point cloud cache, odometer cumulative reading and optimizer status; at the same time, scan data that has not yet formed a complete keyframe is written to a temporary cache.

10. The method for segmented continuous mapping of coal mine roadways supporting long-term operation according to claim 9, characterized in that, when restarting after a power outage, the data processing unit first searches for the most recent valid snapshot, prioritizes loading the newest version, and if the verification fails, it rolls back to the backup; then it reads the odometer increments before and after the power outage, adds the temporary cache to the end of the trajectory, and continues mapping.

Images