An explosion-proof gunite robot multi-sensor fusion perception method and system and storage medium

Abstract

This invention discloses a multi-sensor fusion perception method, system, and storage medium for an explosion-proof shotcrete robot. The invention first acquires coal mine roadway structure and raw image data, matching the working space to determine the region of interest; then, it collects 3D point cloud data of this region, and constructs a complete 3D point cloud model of the roadway through rapid indexing, filtering, registration, and stitching; next, it preprocesses the image and registers it with the point cloud, using image recognition to correct obstacle positions; subsequently, it extracts key features from the point cloud to establish a topological network containing obstacles; finally, it combines robot inertial measurement and odometry data, using Kalman filtering to calibrate the pose, completing the fusion perception. This invention significantly improves the accuracy of obstacle localization in complex roadway environments and reduces the risk of robot collisions.

Description

Technical Field

[0001] This invention relates to the field of robot environmental perception, and more specifically to a multi-sensor fusion perception method, system, and storage medium for an explosion-proof shotcrete robot. Background Technology

[0002] Existing shotcrete robots' perception systems mostly rely on single sensors. In high-dust environments, the laser beam of lidar is easily scattered, leading to a surge in point cloud noise, reduced accuracy in boundary feature extraction, and an inability to identify dynamic obstacles. While high-definition visible light cameras can assist in observing shotcrete effects and equipment status, their image clarity is reduced due to the low-light environment underground and dust adhering to the lens, and they cannot directly obtain target distance information. Ultrasonic and level sensors are mostly used for short-range obstacle avoidance, with narrow detection ranges, susceptibility to environmental sound wave interference, and inability to cover the entire working space of the shotcrete robot, nor can they provide overall geometric features of the tunnel. Inertial measurement units on some robots accumulate errors underground due to metal structures and electromagnetic interference, resulting in significant positioning deviations after long-term operation. Moreover, existing sensor data is mostly processed independently without effective fusion. LiDAR point cloud data and visual image feature information are disconnected, and obstacle avoidance sensor alarm signals lack linkage with robotic arm motion control, which can easily lead to "perception blind spots" or "false shutdowns." In addition, coal mines have explosion-proof requirements for sensors, and conventional industrial sensors cannot meet explosion-proof and intrinsically safe design standards. Dedicated explosion-proof sensors also have signal transmission delay and data compatibility issues, further reducing the real-time performance and reliability of the sensing system.

[0003] Current perception solutions for shotcrete robots are no longer sufficient to meet the comprehensive requirements of explosion protection, adaptation to complex environments, high-precision perception, and real-time control in coal mines. To ensure that shotcrete robots can operate fully automatically, safely, and efficiently, there is an urgent need for a perception method based on multi-sensor fusion, which integrates multi-source sensor data and combines explosion-proof design with intelligent fusion algorithms to achieve accurate perception of the tunnel environment. Summary of the Invention

[0004] In view of this, the present invention provides a multi-sensor fusion sensing method, system and storage medium for an explosion-proof shotcrete robot to solve the problems existing in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A multi-sensor fusion sensing method for an explosion-proof shotcrete robot includes the following steps:

[0007] Acquire structural data and raw image data of coal mine roadways, match the structural data of coal mine roadways with the working space of shotcrete robots, and select the matched areas as regions of interest.

[0008] Collect 3D point cloud data corresponding to the region of interest, and obtain a complete 3D point cloud model of the tunnel by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud.

[0009] The original image data is preprocessed and spatially registered with the three-dimensional point cloud model of the tunnel. The correspondence between the image and the point cloud is established through feature matching. The position information of obstacles in the point cloud model is corrected using the image recognition results.

[0010] Extract the tunnel wall normal vectors, boundary feature points, and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topological network containing the tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles.

[0011] Based on the topology network, and combined with the inertial measurement unit data and odometer data carried by the shotcrete robot itself, the real-time position and attitude of the robot are dynamically calibrated by the Kalman filter algorithm to complete the robot's fusion perception.

[0012] Optionally, it also includes real-time scanning of environmental changes within the area of ​​interest, comparing and analyzing the obtained dynamic data with the established topology network to determine whether there are new obstacles or the original obstacles have shifted in position. If so, the topology network and the three-dimensional point cloud model of the tunnel are updated.

[0013] Optionally, this also includes collecting real-time operational data of the shotcrete area during the operation of the shotcrete robot, comparing the operational data with the planned parameters in real time, and adjusting the robot's working posture and shotcrete parameters if deviations are found; after the operation is completed, a comprehensive scan of the shotcrete area is performed to obtain the three-dimensional point cloud model and image data after shotcreting, and comparing and evaluating them with the shotcrete operation requirements to determine whether the shotcrete quality meets the standards. If it meets the standards, the perception process is completed; if it does not meet the standards, the reasons are analyzed and a secondary shotcrete plan is formulated. Based on the perception data, the above steps are repeated until the shotcrete quality meets the standards, ultimately achieving accurate perception of the working environment of the coal mine roadway and its own working status by the shotcrete robot.

[0014] Optionally, a topological network is established, including tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles. Specifically, this includes: using boundary feature points as topological nodes, calculating the spatial distance between adjacent boundary feature points and using it as node connection weights; using shape decomposition points as partition markers of the topological network, dividing the three-dimensional space of the tunnel into several continuous working sub-regions based on the shape decomposition points; constructing a regional orientation feature matrix through tunnel wall normal vectors, using matrix similarity as the correlation coefficient between adjacent sub-regions; using obstacles as special nodes in the topological network, recording the shortest distance and spatial orientation relationship between obstacles and surrounding boundary feature points, forming constraint edges for the influence range of obstacles; finally, through node connection weights, sub-region correlation coefficients, and obstacle constraint edges, a topological network containing spatial positional relationships, regional divisions, and obstacle constraints is constructed to achieve a structured description of the tunnel environment.

[0015] Optionally, the raw image data may be preprocessed, specifically including the following steps:

[0016] A composite filtering method combining Gaussian filtering and bilateral filtering is adopted. First, Gaussian filtering is used to perform initial noise reduction on the original image to remove Gaussian noise. Then, bilateral filtering is used to further process the image after initial noise reduction. While preserving the image edge information, salt-and-pepper noise and residual fine noise are removed to obtain the image after preliminary noise reduction.

[0017] The weighted average method is adopted, and weight coefficients are set for the sensitivity of different color channels. The pixel values ​​of the red, green and blue channels of the color image are converted into grayscale image pixel values ​​according to the weights to form a grayscale image, thereby reducing the amount of data in subsequent processing.

[0018] First, a histogram equalization algorithm is used to globally enhance the grayscale image, making the grayscale distribution more uniform and improving the overall contrast. Then, the Laplacian operator is used to perform convolution operations on the globally enhanced image to highlight the image edges and details and enhance the image clarity. Finally, the preprocessed image data is obtained.

[0019] Optionally, the structural data of the coal mine roadway can be matched with the working space of the shotcrete robot. Specifically, by using coordinate mapping, the static and dynamic structural data of the coal mine roadway can be uniformly converted to the coordinate system of the shotcrete robot working space model, and the coordinate correspondence between the roadway structural data and the working space can be established to complete the spatial matching.

[0020] A multi-sensor fusion sensing system for an explosion-proof shotcrete robot includes:

[0021] Region of Interest (ROI) matching module: This module is used to acquire structural data and raw image data of coal mine roadways, match the structural data of coal mine roadways with the working space of shotcrete robots, and take the matched area as the region of interest.

[0022] The tunnel 3D point cloud model construction module is used to collect 3D point cloud data corresponding to the region of interest, and obtain a complete tunnel 3D point cloud model by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud.

[0023] Obstacle localization module: It is used to preprocess the raw image data and spatially register it with the 3D point cloud model of the tunnel. It establishes the correspondence between the image and the point cloud through feature matching and uses the image recognition results to correct the position information of obstacles in the point cloud model.

[0024] Topology network establishment module: used to extract tunnel wall normal vectors, boundary feature points and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topology network containing tunnel wall normal vectors, boundary feature points, shape decomposition points and obstacles;

[0025] Robot Fusion Perception Module: Based on the topology network, and combined with the inertial measurement unit data and odometry data carried by the shotcrete robot itself, the module uses the Kalman filter algorithm to dynamically calibrate the robot's real-time position and attitude, thus completing the robot's fusion perception.

[0026] A computer storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any one of the methods for a multi-sensor fusion sensing method for an explosion-proof shotcrete robot.

[0027] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a multi-sensor fusion sensing method, system and storage medium for an explosion-proof shotcrete robot, which has the following beneficial effects:

[0028] 1. By matching coal mine roadway structure data with the robot's operating space, the "region of interest" is accurately delineated, avoiding invalid perception of non-operating areas and reducing the redundancy of subsequent data collection and processing. At the same time, the system focuses on collecting 3D point clouds in key areas, and combines fast indexing, filtering, registration and stitching to effectively remove noise points and invalid points, ensuring the integrity and accuracy of the roadway 3D point cloud model and providing high-quality basic data for subsequent perception.

[0029] 2. By preprocessing images and registering them in 3D point cloud space, and using feature matching to establish the correspondence between the two, we can utilize the advantages of the 3D spatial location of point clouds and rely on the clear discrimination capability of image recognition to correct the position information of obstacles in the point cloud model. This solves the perception limitations of a single sensor, greatly improves the accuracy of obstacle positioning in complex alleyway environments, and reduces the risk of robot collisions.

[0030] 3. Extract the tunnel wall normal vector, boundary feature points, and shape decomposition points from the point cloud model, and integrate obstacle information to construct a topological network. Transform the scattered spatial features into structured relational data, intuitively presenting the tunnel wall morphology, boundary range, and obstacle distribution. Provide a clear "environmental map" reference for the robot to plan the spraying path and adjust its working posture, avoiding operational deviations caused by blurred environmental features. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the method flow provided by the present invention;

[0033] Figure 2 This is a schematic diagram of the system structure provided by the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] This invention discloses a multi-sensor fusion sensing method for an explosion-proof shotcrete robot, such as... Figure 1 As shown, it includes the following steps:

[0036] Step 1: Obtain structural data and raw image data of the coal mine roadway, match the structural data of the coal mine roadway with the working space of the shotcrete robot, and select the matched area as the region of interest;

[0037] Step 2: Collect 3D point cloud data corresponding to the region of interest, and obtain a complete 3D point cloud model of the tunnel by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud.

[0038] Step 3: Preprocess the original image data and spatially register it with the 3D point cloud model of the tunnel. Establish the correspondence between the image and the point cloud through feature matching, and use the image recognition results to correct the position information of obstacles in the point cloud model.

[0039] Step 4: Extract the tunnel wall normal vectors, boundary feature points, and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topological network containing the tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles.

[0040] Step 5: Based on the topology network, and combined with the inertial measurement unit data and odometry data carried by the shotcrete robot itself, the real-time position and attitude of the robot are dynamically calibrated using the Kalman filter algorithm to complete the robot's fusion perception.

[0041] Furthermore, in step one, the original image data is preprocessed, specifically including the following steps:

[0042] A composite filtering method combining Gaussian filtering and bilateral filtering is adopted. First, Gaussian filtering is used to perform initial noise reduction on the original image to remove Gaussian noise. Then, bilateral filtering is used to further process the image after initial noise reduction. While preserving the image edge information, salt-and-pepper noise and residual fine noise are removed to obtain the image after preliminary noise reduction.

[0043] The weighted average method is adopted, and weight coefficients are set for the sensitivity of different color channels. The pixel values ​​of the red, green and blue channels of the color image are converted into grayscale image pixel values ​​according to the weights to form a grayscale image, thereby reducing the amount of data in subsequent processing.

[0044] First, a histogram equalization algorithm is used to globally enhance the grayscale image, making the grayscale distribution more uniform and improving the overall contrast. Then, the Laplacian operator is used to perform convolution operations on the globally enhanced image to highlight the image edges and details and enhance the image clarity. Finally, the preprocessed image data is obtained.

[0045] Furthermore, the structural data of the coal mine roadway is matched with the working space of the shotcrete robot, specifically:

[0046] First, the acquired coal mine roadway structure data is classified and processed into static structure data and dynamic structure data. The static structure data includes the design length, width, height, wall slope, and location and dimensions of the fixed support structure of the roadway. The dynamic structure data includes the location range of temporary material storage, the distribution area of ​​temporary construction equipment, and the travel trajectory range of passing transportation equipment.

[0047] Based on the mechanical structure parameters of the shotcrete robot (including robot body size, maximum extension length of shotcrete arm, range of shotcrete arm rotation angle and range of shotcrete operation blind zone) and operation performance parameters (including maximum shotcrete coverage area and single operation movement distance limit), a three-dimensional operation space model of the shotcrete robot is constructed to clarify the effective coverage boundary and non-operational area of ​​the operation space.

[0048] By using coordinate mapping, the static and dynamic structural data of coal mine roadways are uniformly transformed into the coordinate system of the shotcrete robot's working space model, thus establishing the coordinate correspondence between roadway structural data and working space.

[0049] Set matching rules: When the wall area and the gap area of ​​the fixed support structure in the static structure data of the roadway are completely within the effective coverage boundary of the shotcrete robot's working space, and the temporary material and temporary equipment areas in the dynamic structure data do not overlap with the non-operable areas of the working space, and the area maintains a preset safe distance from the passing trajectory range of the transport equipment, the roadway area is determined to match the shotcrete robot's working space.

[0050] The entire structural data area of ​​the coal mine roadway is traversed, and all areas that meet the matching rules are selected. These areas are then merged to form a continuous region of interest. If there are discontinuous matching areas and the distance between adjacent areas is less than the single operation distance of the shotcrete robot, the transition area between the two areas is supplemented through path planning to ensure the integrity and operability of the region of interest.

[0051] To further improve the accuracy of perception, it also includes real-time scanning of environmental changes within the area of ​​interest, comparing and analyzing the dynamic data obtained from the scan with the established topology network to determine whether there are new obstacles or whether the positions of existing obstacles have shifted. If so, the topology network and the three-dimensional point cloud model of the tunnel are updated.

[0052] In addition, the system includes real-time collection of operational data in the shotcrete area during the operation of the shotcrete robot, real-time comparison of the operational data with the planning parameters, and adjustment of the robot's working posture and shotcrete parameters if deviations are found. After the operation is completed, the shotcrete area is fully scanned to obtain the three-dimensional point cloud model and image data after shotcreting, and the data is compared and evaluated with the shotcrete operation requirements to determine whether the shotcrete quality meets the standards. If it meets the standards, the perception process is completed; if it does not meet the standards, the reasons are analyzed and a secondary shotcrete plan is formulated. Based on the perception data, the above steps are repeated until the shotcrete quality meets the standards, ultimately achieving accurate perception of the working environment of the coal mine roadway and the robot's own working status by the shotcrete robot.

[0053] Furthermore, in step four, a topological network is established, including tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles. Specifically, this includes: using boundary feature points as topological nodes, calculating the spatial distance between adjacent boundary feature points and using it as node connection weights; using shape decomposition points as partition markers of the topological network, dividing the tunnel's three-dimensional space into several continuous working sub-regions based on the shape decomposition points; constructing a regional orientation feature matrix using tunnel wall normal vectors, and using matrix similarity as the correlation coefficient between adjacent sub-regions; using obstacles as special nodes in the topological network, recording the shortest distance and spatial orientation relationship between obstacles and surrounding boundary feature points, forming constraint edges for the obstacle's influence range; and finally, constructing a topological network containing spatial positional relationships, regional divisions, and obstacle constraints through node connection weights, sub-region correlation coefficients, and obstacle constraint edges, thereby achieving a structured description of the tunnel environment.

[0054] This embodiment addresses the problem of blurred image edges caused by dim lighting and dust interference in coal mine roadways. Based on a dynamic edge detection algorithm using local image contrast, it generates a comprehensive edge map and calculates a sharpness score to accurately extract the contour features of roadway walls and obstacles. The specific steps are as follows:

[0055] Dynamic edge detection algorithm selection: The preprocessed image is traversed using a sliding window of preset size. The ratio of the difference between the maximum and minimum gray values ​​of pixels within the window to their sum is calculated as the local contrast. Multiple local contrast threshold ranges are set, and different edge detection algorithms are selected for different ranges. For example, when the local contrast is higher than a certain high threshold, a combination of the Canny algorithm and the Sobel operator is used, leveraging the edge localization advantage of the Canny algorithm and the sensitivity of the Sobel operator to edges in different directions to enhance edge details in high-contrast areas. When the local contrast is in the middle threshold range, the Gaussian-Laplacian (LoG) operator is used to calculate the edge response using Gaussian kernels of different scales to capture the contours of medium-contrast areas. When the local contrast is lower than a certain low threshold, a phase-consistent edge detection algorithm is selected. By calculating the local phase consistency of the image in multiple directions, edges are effectively extracted in low-contrast environments, avoiding edge loss.

[0056] Generate a comprehensive edge map: For the initial edge maps output by each algorithm, determine the weight of each pixel in the comprehensive edge map based on the proportion of the edge gradient magnitude at each pixel position to the maximum gradient magnitude of the entire image; perform weighted fusion on different initial edge maps of the same image according to the weights, and perform non-maximum suppression operation on the image after fusion to eliminate redundant edges; then use morphological closing operation to process the image with structuring elements of appropriate size to connect broken edges, and finally generate a comprehensive edge map containing the weight information of each pixel, which fully presents the outline of the alley wall and obstacles.

[0057] Calculate the sharpness score: Calculate the sum of the gradient magnitudes of all pixels in the comprehensive edge map and divide it by the total number of pixels to obtain the average gradient magnitude; Pre-set the minimum and maximum values ​​of the average gradient magnitude to correspond to blurred and high-sharp images, respectively, and normalize the calculated average gradient magnitude to the [0,1] interval as the image sharpness score; If the original image data comes from multiple perspectives (such as images captured by multiple cameras of a shotcrete robot), take the minimum image sharpness score of each perspective as the final sharpness score of the original image data, ensuring that subsequent processing is based on image data with acceptable edge quality, and providing reliable edge information for spatial registration of the 3D point cloud model of the tunnel and the image, as well as obstacle position correction.

[0058] and Figure 1 Corresponding to the method shown, this invention also discloses a multi-sensor fusion sensing system for an explosion-proof shotcrete robot for... Figure 1 The implementation of the method, specifically the structure is as follows: Figure 2 As shown, it includes:

[0059] Region of Interest (ROI) matching module: This module is used to acquire structural data and raw image data of coal mine roadways, match the structural data of coal mine roadways with the working space of shotcrete robots, and take the matched area as the region of interest.

[0060] The tunnel 3D point cloud model construction module is used to collect 3D point cloud data corresponding to the region of interest, and obtain a complete tunnel 3D point cloud model by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud.

[0061] Obstacle localization module: It is used to preprocess the raw image data and spatially register it with the 3D point cloud model of the tunnel. It establishes the correspondence between the image and the point cloud through feature matching and uses the image recognition results to correct the position information of obstacles in the point cloud model.

[0062] Topology network establishment module: used to extract tunnel wall normal vectors, boundary feature points and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topology network containing tunnel wall normal vectors, boundary feature points, shape decomposition points and obstacles;

[0063] Robot Fusion Perception Module: Based on the topology network, and combined with the inertial measurement unit data and odometry data carried by the shotcrete robot itself, the module uses the Kalman filter algorithm to dynamically calibrate the robot's real-time position and attitude, thus completing the robot's fusion perception.

[0064] Finally, this embodiment discloses a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of any one of the methods for a multi-sensor fusion perception method for an explosion-proof shotcrete robot.

[0065] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0066] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-sensor fusion sensing method for an explosion-proof shotcrete robot, characterized in that, Includes the following steps: Acquire structural data and raw image data of coal mine roadways, match the structural data of coal mine roadways with the working space of shotcrete robots, and select the matched areas as regions of interest. Collect 3D point cloud data corresponding to the region of interest, and obtain a complete 3D point cloud model of the tunnel by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud. The original image data is preprocessed and spatially registered with the three-dimensional point cloud model of the tunnel. The correspondence between the image and the point cloud is established through feature matching. The position information of obstacles in the point cloud model is corrected using the image recognition results. Extract the tunnel wall normal vectors, boundary feature points, and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topological network containing the tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles. Based on the topology network, and combined with the inertial measurement unit data and odometer data carried by the shotcrete robot itself, the robot's real-time position and attitude are dynamically calibrated using the Kalman filter algorithm to complete the robot's fusion perception.

2. The multi-sensor fusion sensing method for an explosion-proof shotcrete robot according to claim 1, characterized in that, It also includes real-time scanning of environmental changes within the area of ​​interest, comparing and analyzing the dynamic data obtained from the scan with the established topology network to determine whether there are new obstacles or whether the positions of existing obstacles have shifted. If so, the topology network and the three-dimensional point cloud model of the tunnel are updated.

3. The multi-sensor fusion sensing method for an explosion-proof shotcrete robot according to claim 1, characterized in that, It also includes collecting real-time operational data of the shotcrete area during the operation of the shotcrete robot, comparing the operational data with the planning parameters in real time, and adjusting the robot's operating posture and shotcrete parameters if a deviation is found. After the operation is completed, a comprehensive scan of the shotcrete area is performed to obtain the three-dimensional point cloud model and image data after shotcrete. The data is then compared and evaluated with the shotcrete operation requirements to determine whether the shotcrete quality meets the standards. If it meets the standards, the perception process is completed. If it does not meet the standards, the reasons are analyzed and a secondary shotcrete plan is formulated. Based on the perception data, the above steps are repeated until the shotcrete quality meets the standards. Ultimately, the shotcrete robot achieves accurate perception of the coal mine roadway working environment and its own working status.

4. The multi-sensor fusion sensing method for an explosion-proof shotcrete robot according to claim 1, characterized in that, A topological network is established, comprising tunnel wall normal vectors, boundary feature points, shape decomposition points, and obstacles. Specifically, this includes: using boundary feature points as topological nodes, calculating the spatial distance between adjacent boundary feature points and using it as node connection weights; using shape decomposition points as partition markers of the topological network, dividing the three-dimensional space of the tunnel into several continuous working sub-regions based on the shape decomposition points; constructing regional orientation feature matrices using tunnel wall normal vectors, and using matrix similarity as the correlation coefficient between adjacent sub-regions; treating obstacles as special nodes in the topological network, recording the shortest distance and spatial orientation relationship between obstacles and surrounding boundary feature points, forming constraint edges for the obstacle's influence range; and finally, constructing a topological network containing spatial positional relationships, regional divisions, and obstacle constraints through node connection weights, sub-region correlation coefficients, and obstacle constraint edges, thereby achieving a structured description of the tunnel environment.

5. The multi-sensor fusion sensing method for an explosion-proof shotcrete robot according to claim 1, characterized in that, Preprocessing the raw image data includes the following steps: A composite filtering method combining Gaussian filtering and bilateral filtering is adopted. First, Gaussian filtering is used to perform initial noise reduction on the original image to remove Gaussian noise. Then, bilateral filtering is used to further process the image after initial noise reduction. While preserving the image edge information, salt-and-pepper noise and residual fine noise are removed to obtain the image after preliminary noise reduction. The weighted average method is adopted, and weight coefficients are set for the sensitivity of different color channels. The pixel values ​​of the red, green and blue channels of the color image are converted into grayscale image pixel values ​​according to the weights to form a grayscale image, thereby reducing the amount of data in subsequent processing. First, a histogram equalization algorithm is used to globally enhance the grayscale image, making the grayscale distribution more uniform and improving the overall contrast. Then, the Laplacian operator is used to perform convolution operations on the globally enhanced image to highlight the image edges and details and enhance the image clarity. Finally, the preprocessed image data is obtained.

6. The multi-sensor fusion sensing method for an explosion-proof shotcrete robot according to claim 1, characterized in that, Matching the structural data of coal mine roadways with the working space of shotcrete robots involves: using coordinate mapping to uniformly transform the static and dynamic structural data of coal mine roadways into the coordinate system of the shotcrete robot's working space model, establishing the coordinate correspondence between the roadway structural data and the working space, and completing the spatial matching.

7. A multi-sensor fusion sensing system for an explosion-proof shotcrete robot, characterized in that, include: Region of Interest (ROI) matching module: This module is used to acquire structural data and raw image data of coal mine roadways, match the structural data of coal mine roadways with the working space of shotcrete robots, and take the matched area as the region of interest. The tunnel 3D point cloud model construction module is used to collect 3D point cloud data corresponding to the region of interest, and obtain a complete tunnel 3D point cloud model by establishing a fast index, filtering to remove invalid and noise points, and registering and stitching the point cloud. Obstacle localization module: It is used to preprocess the raw image data and spatially register it with the 3D point cloud model of the tunnel. It establishes the correspondence between the image and the point cloud through feature matching and uses the image recognition results to correct the position information of obstacles in the point cloud model. Topology network establishment module: used to extract tunnel wall normal vectors, boundary feature points and shape decomposition points from the 3D point cloud model of the tunnel, and establish a topology network containing tunnel wall normal vectors, boundary feature points, shape decomposition points and obstacles; Robot Fusion Perception Module: Based on the topology network, and combined with the inertial measurement unit data and odometry data carried by the shotcrete robot itself, the module uses the Kalman filter algorithm to dynamically calibrate the robot's real-time position and attitude, thus completing the robot's fusion perception.

8. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of a multi-sensor fusion sensing method for an explosion-proof shotcrete robot as described in any one of claims 1-6.

Images