A method and system for unmanned aerial vehicle path planning based on laser scanning

Abstract

The present application relates to the technical field of unmanned aerial vehicle path planning, and particularly relates to an unmanned aerial vehicle path planning method and system based on laser scanning. The method comprises the following steps: real-time scanning of a tunnel environment by a laser scanner to obtain original radial profile point cloud data; analysis of a structure stability index and a dangerous rock potential energy threat value based on the original radial profile point cloud data to determine a risk attribution area of a current unmanned aerial vehicle flight path; determination of whether a high-risk detour area exists according to the risk attribution area, and if so, dynamic planning of a risk detour path to adjust a single pre-scanning voyage and generate an intelligent safe execution air route; instructing the unmanned aerial vehicle to fly along the intelligent safe execution air route to its endpoint, and taking the endpoint as a starting point for the next round of path planning until the iterative exploration of the target underground mine space is completed. The present application quantifies the structure stability and potential energy threat of surrounding rock in real time to achieve safe path planning for active avoidance of surrounding rock instability.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning technology, and in particular to a UAV path planning method and system based on laser scanning. Background Technology

[0002] In long-term mined tunnels, geologically complex mining areas, or regions that have experienced ground pressure activity, the structural stability of the environment is not uniform. Due to the combined effects of stress release, weathering, and groundwater erosion, the tunnel roof and sidewalls typically develop numerous joints and interconnected fissures. These fissures gradually expand, eventually causing parts of the rock mass to separate from the parent rock (main rock mass), forming "dangerous rock masses" or "suspended bodies" suspended above the tunnel by only a few connection points. These dangerous rock masses are among the most critical safety hazards in underground mines. Their state is extremely unstable, and they can fall at any time due to minor disturbances (such as equipment vibration, airflow impact, or even downwash generated by drone rotors), causing serious safety accidents. Existing drone path planning methods typically treat the scanned environment as a geometrically stable and homogeneous whole. Their core safety logic mainly focuses on "obstacle avoidance" based on physical distance, severely lacking the ability to perceive and assess the structural stability of the surrounding rock (the rock mass around the tunnel), thus failing to prevent safety accidents caused by rockfalls. Summary of the Invention

[0003] Based on this, the present invention provides a method and system for UAV path planning based on laser scanning to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a UAV path planning method based on laser scanning includes the following steps:

[0005] Step S1: Set a single pre-scan flight path, instruct the UAV to fly along the flight path, and use a laser scanner to scan the tunnel environment in real time to obtain the original radial profile point cloud data;

[0006] Step S2: Separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path;

[0007] Step S3: Determine whether there is a high-risk detour area based on the risk attribution area. If so, dynamically plan the risk detour route to adjust the single pre-scan flight path and generate an intelligent and safe execution route.

[0008] Step S4: Instruct the UAV to fly along the intelligent and safe execution route to its destination, and use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

[0009] The present invention also provides a laser scanning-based UAV path planning system, which executes the laser scanning-based UAV path planning method described above. The laser scanning-based UAV path planning system includes:

[0010] The flight range scanning perception module is used to set a single pre-scan flight range, instruct the UAV to fly along the flight range, and use a laser scanner to scan the tunnel environment in real time to obtain raw radial profile point cloud data.

[0011] The real-time risk assessment module is used to separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path;

[0012] The intelligent route generation module is used to determine whether there are high-risk detour areas based on the risk attribution area. If so, it dynamically plans risk detour routes to adjust the single pre-scan route and generate intelligent and safe execution routes.

[0013] The navigation control and iteration module is used to instruct the UAV to fly along a smart and safe execution route to its destination, and use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

[0014] The beneficial effects of this invention are as follows:

[0015] On the one hand, this invention achieves a leap in the core capability of UAVs from traditional geometric obstacle avoidance to perception of surrounding rock structure stability by deeply analyzing the original radial profile point cloud data to separate and quantify free point clusters. Existing technologies treat all rock surfaces as homogeneous obstacle boundaries, while this invention, by separating potential "dangerous rock masses" (i.e., free point clusters) and further calculating their structural stability index and potential rock energy threat value, can accurately identify and quantify weak areas that pose a potential fall threat to flight safety. Furthermore, this risk assessment mechanism based on structural stability enables UAVs to distinguish between "stable rock walls" and "suspended dangerous rocks," fundamentally solving the major safety defect of traditional path planning methods that cannot foresee and avoid the risk of rockfall, and greatly improving the survivability and operational safety of UAVs in complex geological environments.

[0016] On the other hand, this invention dynamically generates intelligent and safe flight paths based on risk assessment results, realizing a path planning mode that shifts from passive obstacle avoidance to proactive risk mitigation. When a high-risk detour area is identified, this invention does not simply mark it as a no-fly zone. Instead, it constructs a risk potential field based on the structural stability index and the threat value of the unstable rock mass, dynamically planning an optimal risk detour path that comprehensively considers safety distance and detour costs. This dynamic adjustment and replanning mechanism ensures that the UAV can actively move away from the hazard source, avoiding the triggering of rockfalls due to its own flight disturbances (such as downwash). Simultaneously, this intelligent flight path not only ensures flight safety but can also be optimized for performing multi-angle, high-density supplementary scanning tasks on identified unstable rock masses, thereby obtaining their precise geometric and attitude information while ensuring safety. This provides unprecedented high-value data support for subsequent human intervention and disaster management.

[0017] On the other hand, after each intelligent and safe flight path completed, the drone uses its current endpoint as a new starting point, repeating the "scan-evaluate-plan-fly" cycle. This allows it to gradually and safely penetrate unknown and potentially dangerous areas. This iterative exploration mode ensures that the drone always operates within its latest perceived safety boundaries, effectively avoiding the problem of entering unknown and dangerous areas due to long-distance planning in one go. This significantly enhances the robustness and success rate of the entire exploration mission, ultimately achieving comprehensive and safe autonomous exploration of the target underground mine space. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the UAV path planning method based on laser scanning according to the present invention.

[0019] Figure 2 This is a schematic diagram of the module of the UAV path planning system based on laser scanning of the present invention;

[0020] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a path planning method for unmanned aerial vehicles (UAVs) based on laser scanning, comprising the following steps:

[0022] Step S1: Set a single pre-scan flight path, instruct the UAV to fly along the flight path, and use a laser scanner to scan the tunnel environment in real time to obtain the original radial profile point cloud data;

[0023] In this embodiment of the invention, when the UAV begins an exploration mission, it first uses sonar to conduct forward-looking detection to determine a safe initial flight path without obvious obstacles. While the UAV flies along this path, its onboard laser scanner will continuously scan the surrounding environment 360 degrees and collect high-density three-dimensional point cloud data.

[0024] In one implementation of this invention, assuming the drone is located at the entrance of a passageway, the broadband sonar detector on board the drone scans a 120° fan-shaped area in front, with a maximum detection distance of 50 meters, acquiring a set of initial acoustic echo data. Analyzing the echo data, the maximum passable distance is identified as 42 meters. An 80% safety margin is set, and the length of a single pre-scan flight is calculated to be 42 meters. 80% = 33.6 meters. Based on the geometric center of the echo profile, a central axis with a length of 33.6 meters is generated, pointing towards the depth of the alley, as the single pre-scan range. The UAV is instructed to fly at a constant speed along this 33.6-meter range. During this process, its onboard dual-axis rotating laser scanner continuously performs radial scans on a plane orthogonal to the flight direction at a frequency of 20Hz. Each rotation of the scanner generates a radial profile containing approximately 1800 laser points, forming the original radial profile point cloud data.

[0025] Step S2: Separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path;

[0026] In one implementation of this invention, the acquired original radial profile point cloud data is voxelized, with a grid size of 10cm x 10cm x 10cm. For each voxel, the positional eccentricity between the geometric center of all laser pulse points and the corresponding voxel geometric center, as well as the standard deviation of the echo intensity values ​​of all points, are calculated to obtain the intensity fluctuation value. A geometric clustering threshold of 5cm and a material homogeneity threshold of 15 are set. When the positional eccentricity of a voxel is less than 5cm and the intensity fluctuation value is less than 15, the voxel is marked as a high-confidence voxel. The set of pulse point centroids within all high-confidence voxels constitutes the structural confidence point cloud. A region growing algorithm is performed on the structural confidence point cloud to identify and extract the largest connected point cloud set, which is then identified as the main rock mass. Subsequently, a Boolean difference operation is performed between the structural confidence point cloud and the main rock mass to obtain the initial suspended body. Euclidean distance clustering analysis is performed on the initial suspended body, with a clustering distance threshold set to 50cm. After clustering, small clusters (noise) with fewer than 50 points are filtered out, and the remaining point cloud clusters are the free point clusters. Assume three free point clusters are identified, labeled C1, C2, and C3. Taking C1 as an example, assuming its structural stability index is 1.4 and its rockfall threat value is 15.68, and setting the high-risk screening threshold to 1.5, since C1's structural stability index of 1.4 is less than 1.5, and its rockfall threat value of 15.68 is relatively high, the area where C1 is located and its surrounding space are marked as a high-risk detour area, while other areas are considered normal flight areas.

[0027] Step S3: Determine whether there is a high-risk detour area based on the risk attribution area. If so, dynamically plan the risk detour route to adjust the single pre-scan flight path and generate an intelligent and safe execution route.

[0028] In this embodiment of the invention, once a high-risk area is detected, the original straight-line pre-scanning route will be abandoned. For example, for the high-risk free point cluster C1 identified in the previous step, a spatial influence radius of 10 meters is set with its centroid as the center. Within this radius, a Gaussian potential field function is constructed, the peak intensity A of which is proportional to the potential energy threat value of the dangerous rock (…). Its attenuation width σ is proportional to the reciprocal of the structural stability index ( The potential field data is overlaid onto a 3D spatial raster map to form flight risk potential field data. Within this data, the starting point (current UAV position) and ending point (originally planned at 33.6 meters) of the original "single pre-scan flight path" are used as the starting and target points for path searching. The path cost function is defined as: Cost = w1. Path length + w2 Total value of traversing risk potential field +w3 Energy consumption is estimated, where weights w1, w2, and w3 are adjustable. A fast expanding random tree algorithm is used to search within the risk potential field. During expansion, the algorithm tends to avoid areas with high potential values ​​(because this significantly increases costs). When the search reaches the maximum number of iterations (e.g., 5000) or a path with a cost lower than a preset threshold is found, the feasible path with the lowest current cost is output as a risk detour path. This path gracefully bypasses the high-risk area where C1 is located and eventually reaches the original target point. This new path is the intelligent and safe execution route.

[0029] Step S4: Instruct the UAV to fly along the intelligent and safe execution route to its destination, and use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

[0030] In this embodiment of the invention, the generated intelligent safe execution route (represented as a series of three-dimensional spatial waypoints) is sent to the UAV's flight control system. The flight control system drives the UAV to fly precisely along this safe path. When the UAV flies to the end of the intelligent safe execution route (a new position at a depth of 33.6 meters) and hovers stably, it marks the completion of this exploration cycle. The current position of the UAV is immediately used as the starting point for the next round of path planning. Subsequently, the entire process returns to step S1: the UAV performs forward-looking sonar detection again from the new position, sets a new "single pre-scan range", and repeats the complete closed loop of "scan-evaluation-planning-flight". This iterative process will continue, advancing safely into the depth of the tunnel segment by segment. When the sonar detects the end of the tunnel, or when the total exploration mileage reaches a preset value, the mission ends.

[0031] Preferably, step S1 includes the following steps:

[0032] Step S11: Before the UAV enters the target underground mine space, use its onboard broadband sonar detector to perform fan-shaped acoustic detection on the area in front to obtain initial acoustic echo data containing the main contour reflection information.

[0033] Step S12: Based on the farthest echo boundary identified in the initial acoustic echo data, calculate and set a safe forward distance to obtain the single pre-scan range;

[0034] Step S13: Instruct the UAV to fly along the central axis of the single pre-scan flight path to the end of the single pre-scan flight path. During this process, use the dual-axis rotating laser scanner that is orthogonal to the flight direction to perform real-time radial scanning of the roadway sides and top and bottom plates of the target underground mine space to obtain the original radial profile point cloud data.

[0035] In this embodiment of the invention, at the starting point of each iterative exploration, the UAV first performs a forward long-range coarse reconnaissance. The purpose is to quickly assess the accessibility and general outline of the space ahead, providing a basis for setting a safe flight scanning distance. To reduce cost and power consumption, acoustic detection is used in this stage.

[0036] In one implementation of this invention, the UAV is equipped with a remote ultrasonic sensor array consisting of three transmitting / receiving units. This array can detect a fan-shaped area in the horizontal direction from -60° to +60°. The sensor array emits a series of coded ultrasonic pulses in seven directions—-60°, -40°, ..., +40°, +60°—with a 20° step angle. After detection in each direction, the receiving unit captures a series of acoustic signals reflected from the tunnel wall, obstacles, or distant boundaries. These signals are amplified and filtered, and then recorded as initial acoustic echo data containing flight time and signal amplitude. Specifically, for detection in the 0° (directly forward) direction, it is assumed that the received initial acoustic echo data is a time-domain waveform, with a significant peak at t1 = 0.25s and a weaker peak at t2 = 0.28s, indicating two main reflecting surfaces directly in front.

[0037] In one implementation of this invention, the processing unit analyzes the initial acoustic echo data from seven directions one by one. For each direction, a constant threshold detection method is used to identify the effective peak value in the echo signal, and the flight time of the last effective peak value is recorded. .this The time is considered to correspond to the farthest detection distance in that direction.

[0038] In another implementation of this invention, it is assumed that the flight time is calculated in seven detection directions. Given that the speed of sound in air is 0.25 seconds... If the speed is approximately 340 m / s, then the distance to the farthest echo boundary is... The calculation is as follows: It should be noted that the calculation result here is divided by 2 because the sound wave travels a round trip distance.

[0039] Then, based on this farthest echo boundary distance, a preset safety factor is introduced. To calculate the final safe forward distance. The safety factor is set to allow for sufficient reaction and braking distance. Its mathematical expression is:

[0040] ;

[0041] in: It is the safe forward travel distance; It is the distance to the farthest echo boundary identified; It is a dimensionless coefficient between 0 and 1, which is dynamically adjusted according to the current flight speed of the UAV. Its empirical model can be set as follows: ,in Let the drone's speed be (m / s), assuming the drone is currently hovering. ),but .

[0042] In one implementation of this invention, the UAV flight control system receives a single pre-scan range command with a length of 38 meters. The flight control system generates a straight trajectory of 38 meters in length, using the current position of the UAV as the starting point and the direction directly ahead as the central axis. The UAV flies along this axis at a constant speed of 1.0 m / s. Specifically, the rotation axis of the laser scanner (e.g., Velodyne VLP-16) mounted on the UAV maintains a strictly orthogonal relationship with the UAV's flight direction (central axis). As the UAV flies forward, the scanner's 16-line laser beam rotates horizontally 360° at a frequency of 20 Hz, forming a continuously moving "scanning disk" perpendicular to the flight path. This scanning disk cuts through the tunnel space in real time. The laser beam sequentially scans the right side, roof, left side, and floor of the tunnel, then returns to the right side, completing a radial scan. With each scan, each laser emitter returns one or more echo points. The three-dimensional coordinates (based on the scanner's own coordinate system) and echo intensity values ​​of these echo points are combined with the real-time pose information (position and attitude) provided by the UAV's inertial measurement unit (IMU) and transformed into a unified global coordinate system. All these point cloud sets after coordinate transformation together constitute the original radial profile point cloud data for this flight.

[0043] Preferably, before separating the free point clusters based on the original radial profile point cloud data in step S2, the method further includes:

[0044] The original radial profile point cloud data is rasterized in three dimensions to obtain three-dimensional scan point cloud data.

[0045] The 3D scan point cloud data is divided into spatial voxel grids, and then the laser pulse points in the 3D scan point cloud data are assigned to the corresponding spatial voxel grids to obtain voxelized pulse data.

[0046] Based on the voxelized pulse data, the Euclidean distance between the three-dimensional coordinate geometric center of all laser pulse points in each voxel and the corresponding voxel geometric center is calculated to obtain the positional eccentricity.

[0047] Simultaneously, the standard deviation of the echo intensity values ​​of all laser pulse points within each voxel in the voxelized pulse data is calculated to obtain the intensity fluctuation value.

[0048] Determine whether the positional eccentricity is less than the preset geometric aggregation threshold and whether the intensity fluctuation value is less than the preset material uniformity threshold. If so, mark the voxel as a high-confidence voxel.

[0049] Based on high-confidence voxel analysis of the centroids of impulse points within voxels, the set of all centroids is used to construct a structural confidence point cloud.

[0050] In this embodiment of the invention, each acquired data frame is traversed, and the pose information (a 4x4 homogeneous transformation matrix) is used to transform the point cloud in the local coordinate system of the frame to the global coordinate system (usually with the mission starting point as the origin). All the transformed point clouds are combined together to form three-dimensional scan point cloud data covering the entire single pre-scan flight range area.

[0051] In one implementation of this invention, the side length of a voxel is set to 10 centimeters. A three-dimensional voxel grid covering the entire range of the three-dimensional scanned point cloud data is created. Each laser pulse point in the three-dimensional scanned point cloud data is traversed, and its corresponding voxel index (i, j, k) is calculated based on its three-dimensional coordinates (x, y, z). The point is then stored in the voxel with the corresponding index. The set of all non-empty voxels and the laser pulse points they contain constitutes the voxelized pulse data.

[0052] In this embodiment of the invention, the three-dimensional geometric center of all laser pulse points within each voxel is calculated, and simultaneously, the geometric center of the voxel itself is calculated. Then, the Euclidean distance between these two center points is calculated, and this distance value is defined as the positional eccentricity. For a voxel V containing n points, its positional eccentricity is... The calculation process is as follows: Calculate all n laser pulse points within the voxel. The three-dimensional coordinate geometric center :

[0053] ;

[0054] Obtain the geometric center of the voxel V itself. (Its coordinates are directly determined by the voxel index and side length). Calculate the Euclidean distance between these two center points, which is the positional eccentricity. :

[0055] ;

[0056] in: This is the positional eccentricity, expressed in meters. The coordinates of the geometric center of the point cloud within the voxel; The coordinates are the geometric center coordinates of the voxel itself. Specifically, the standard deviation of the echo intensity values ​​of all laser pulse points within each non-empty voxel in the voxelized pulse data is calculated. This standard deviation is defined as the intensity fluctuation value. The positional eccentricity is then compared with a preset geometric clustering threshold, and simultaneously, the intensity fluctuation value is compared with a preset material homogeneity threshold. It should be noted that the geometric clustering threshold is set to 5 cm in this case, based on the ranging accuracy of the laser scanner and the general smoothness of the tunnel surface; the material homogeneity threshold is set to 15, based on prior knowledge of the statistical distribution of echo intensity for typical materials such as rock and coal face.

[0057] In this embodiment of the invention, after determining all non-empty voxels, only the data of all voxels marked as high confidence are retained, the centroids of all impulse points inside them (i.e. the average value of the three-dimensional coordinates) are calculated, and the set of all these centroids is used to form a new point cloud dataset that has been purified and downsampled. This dataset is the structural confidence point cloud.

[0058] Preferably, step S2, which involves separating the free point clusters based on the original radial profile point cloud data, includes:

[0059] Perform a region growing operation on the structural credibility point cloud to identify and extract the largest connected point cloud set, and determine this set as the main rock mass;

[0060] Perform Boolean difference operations between the structural credibility point cloud and the main rock mass to separate all points that do not belong to the main rock mass, thus obtaining the initial suspended body;

[0061] Cluster analysis using Euclidean distance is performed on the initial suspended body to segment the scattered point cloud on the initial suspended body into multiple independent point cloud clusters, thus obtaining a set of point cloud clusters;

[0062] Calculate the number of points in each cluster of the point cloud cluster set, and filter out tiny clusters with a number lower than the preset noise point threshold. The remaining cluster set is the free point cluster set.

[0063] In this embodiment of the invention, a region growing algorithm based on normal vector and distance constraints is employed. The point with the lowest Z-coordinate in the structural confidence point cloud is selected as the initial seed point, as this point is highly likely to belong to a stable tunnel floor. Then, an efficient neighborhood search is performed using a kd-tree structure to add neighboring points that are less than 25 cm away from the current region's point cloud and whose normal vector angle is less than 10 degrees to the region. This process is iterated until no new points can be added, thus forming a connected point cloud set. This process is repeated for all unvisited points until all points are classified into a certain set.

[0064] In one implementation of this invention, assuming the original structural confidence point cloud has a total of 509,410 points, and the main rock mass determined in the previous step contains 485,120 points, a Boolean difference operation is performed on the point cloud. This involves removing all points that also exist in the main rock mass set from the total set of structural confidence point clouds. Specifically, the remaining point cloud set obtained after calculating the difference is the initial suspended body, containing 509,410 - 485,120 = 24,290 points.

[0065] In this embodiment of the invention, the initial suspended body contains multiple spatially separated potential unstable rock masses. This step effectively segments these spatially independent entities using a clustering algorithm. It should be noted that Euclidean distance clustering is used because spatial proximity is the most fundamental criterion for determining whether a point cloud belongs to the same physical entity. Euclidean distance clustering analysis is performed on the initial suspended body containing 24,290 points. The key parameter, clustering tolerance, is set to 50 centimeters. This threshold is set because it is significantly greater than the average point spacing of the structural reliability point cloud (approximately 10 centimeters), but less than the generally accepted safe distance between independent unstable rock masses, thus effectively distinguishing different suspended rock blocks. Specifically, after the clustering analysis, the initial suspended body is divided into four independent point cloud clusters, named Cluster 1, Cluster 2, Cluster 3, and Cluster 4, with 15,340, 8,300, 380, and 270 points respectively. The set of these four point cloud clusters constitutes the point cloud cluster set.

[0066] In one implementation of this invention, a noise point threshold of 500 points is set. This threshold is based on engineering experience; an entity consisting of fewer than 500 centroid points (at 10 cm voxel resolution) has a small physical volume and is generally not considered a major safety hazard requiring path avoidance. Specifically, the point cloud cluster set obtained in the previous step is traversed, and the number of points in each cluster is checked. Since cluster 1 (15, 340 points) and cluster 2 (8, 300 points) both have more than 500 points, they are retained. Cluster 3 (380 points) and cluster 4 (270 points) are judged as noise and filtered out because their number of points is less than 500. Therefore, the final remaining cluster set is the free point cluster set containing only cluster 1 and cluster 2.

[0067] Preferably, step S2 involves analyzing the structural stability index of the free point cluster and the potential energy threat value of the unstable rock to determine the risk attribution area of ​​the current UAV flight path, including:

[0068] Search for the set of points whose spatial distance from the cluster of free points to the surface points in the main rock mass is less than a preset connection distance threshold, and define this set of points as the connection point set.

[0069] Based on the analysis of the connection point set, the suspension index and fracture density value of the free point cluster are analyzed;

[0070] The structural stability index is obtained by weighting the suspension index and the crack density value to evaluate the structural stability of the connection area.

[0071] High-risk free point clusters in the free point cluster set are selected based on preset high-risk screening thresholds and structural stability index.

[0072] Calculate the local curvature or surface variation of each point in the high-risk free point cluster, and identify the point set with curvature or surface variation greater than the preset convexity threshold as a convex shape, so as to segment out isolated convex clusters;

[0073] Assess the potential threat value of dangerous rocks based on isolated clusters of protrusions;

[0074] The risk attribution area of ​​the current UAV flight path is determined based on the structural stability index and the potential energy threat value of the dangerous rock; the risk attribution area includes the normal flight area and the high-risk detour area.

[0075] In this embodiment of the invention, for each free point cluster identified in the previous step (taking cluster 1, which contains 15,340 points, as an example), a kd-tree data structure is used to quickly search for the nearest neighbor of each point in cluster 1 within the main rock mass. The preset connection distance threshold is set to 15 cm, which is based on the fact that it is slightly larger than the voxel resolution of the point cloud (10 cm), and can effectively capture physically adjacent point pairs or points connected by minute fractures. Therefore, when it is detected that the Euclidean distance between a point in cluster 1 and its nearest neighbor in the main rock mass is less than 15 cm, this pair of points is simultaneously stored in a temporary set. After traversing all points in cluster 1, this temporary set is determined as the connection point set, assuming that the final connection point set contains 850 points.

[0076] In one implementation of this invention, the suspension index is calculated using the volume of the entire free point cluster and the area of ​​the connecting surface formed by the set of connecting points as input, while the fracture density value is calculated by focusing on the distribution characteristics of the point cloud normal vector in the neighborhood of the connecting point set on one side of the main rock mass.

[0077] In one implementation of this invention, assuming that the suspension index of cluster 1 is 1.5 (the larger the value, the more unstable it is) and the crack density is 0.4 (the larger the value, the more unstable it is), the structural stability index is calculated using the following weighted reciprocal sum model. :

[0078] ;

[0079] in, This is a structural stability index; the smaller the value, the more unstable the structure. For suspension index; This represents the fracture density value. and : These are the weights of the suspension index and the crack density value, respectively, and their sum is 1. Here, based on experience, they are set to 0.6 and 0.4 respectively, to indicate that the suspension morphology plays a more important role in the stability assessment.

[0080] In one implementation of this invention, the structural stability index of each free point cluster calculated in the previous step is compared with a preset high-risk screening threshold. The preset high-risk screening threshold is set to 1.5, which is a critical value determined based on mine safety regulations and ground pressure disaster statistics. If the value is higher than this, the structure is considered to have a high probability of instability.

[0081] In one implementation of this invention, for each point in a high-risk free point cluster, a local quadratic surface is fitted based on its surrounding neighborhood points (e.g., all points within a radius of 0.2 meters), and the principal curvature of that point is calculated. The larger of the two principal curvatures is taken as the local curvature value of that point, which effectively reflects the degree of convexity at the point's location. A preset convexity threshold is set to 0.8, which is used to distinguish sharp, protruding edges or corners on the rock mass from relatively flat surfaces. The set of all points with a local curvature greater than 0.8 is identified as a convex morphology, and these points are then subjected to a small-scale Euclidean distance clustering to segment the largest connected components, forming isolated convex clusters.

[0082] In this embodiment of the invention, once at least one high-risk free point cluster (such as cluster 1 in this example) is identified in the scene, a high-risk detour area is defined based on its spatial location and size. The axis-aligned bounding box of high-risk free point cluster 1 is calculated, and this bounding box is extended outward by 5 meters in each direction as a safety buffer. This expanded three-dimensional space is defined as the high-risk detour area. Therefore, in the entire three-dimensional space covered by the current single pre-scan flight path, all other areas except for this clearly marked high-risk detour area are determined as normal flight areas.

[0083] Preferably, assessing the potential threat value of unstable rocks based on isolated clusters of protrusions includes:

[0084] Construct the minimum convex hull of all outer points in an isolated cluster of protrusions to obtain the dangerous rock protrusion body;

[0085] The number of all non-empty voxel units containing at least one point cloud data within the envelope of an isolated convex cluster is counted, and this number is multiplied by the volume of a single voxel unit to obtain the solid filling product.

[0086] The solid filling volume is compared with the total volume of the unstable rock mass, and the ratio of the degree of development of internal cavities or fissures is calculated as the structural porosity.

[0087] The potential energy threat value of the unstable rock mass is determined based on its volume and structural looseness, and according to the height of the centroid of the isolated cluster of protrusions relative to the roadway floor.

[0088] In this embodiment of the invention, the three-dimensional QuickHull algorithm is used to process the isolated convex cluster containing 1,200 points segmented from the high-risk free point cluster in the previous step. This algorithm can efficiently find a minimal convex polyhedron containing all points. After the algorithm is executed, a closed three-dimensional mesh model composed of 98 triangular facets is generated. By calculating the volume enclosed by this closed mesh, its total volume is found to be 1.5 cubic meters. This geometry and its volume are collectively defined as the dangerous rock convex heap.

[0089] In one implementation of this invention, within the geometric range of the unstable rock mass heap, all voxel units with a side length of 10 cm are counted, and those non-empty voxel units that contain at least one isolated heap cluster point cloud data are identified. Assuming the total number of non-empty voxel units is 11,500 and the volume of a single voxel unit is 0.001 cubic meters, the two are multiplied to obtain the solid filling volume. It should be noted that all voxels within the envelope of the isolated heap cluster (i.e., the unstable rock mass heap) are traversed, and each voxel is checked to see if it contains at least one point cloud data from that cluster. If it does, the voxel is marked as a non-empty voxel unit. The total number of non-empty voxel units is multiplied by the volume of a single voxel unit to obtain a volume value that more realistically reflects the solid portion of the rock mass.

[0090] Specifically, the average three-dimensional coordinates of all points in the isolated cluster of protrusions are calculated to obtain its centroid. Then, the vertical distance from this centroid to the pre-identified roadway floor plane is calculated to obtain its height. The value is 8.5 meters. In one implementation of this invention, the potential energy threat value of the dangerous rock is calculated using the following formula. :

[0091] ;

[0092] in: The potential energy threat value of a dangerous rock is a comprehensive risk indicator. The total volume of the unstable rock mass is represented by its maximum mass. This is a correction factor that takes into account the degree of internal fragmentation. The higher the porosity, the larger the factor, which means that it breaks into more small pieces during the fall and has a wider impact range. The height of the center of mass represents its initial potential energy.

[0093] Preferably, the analysis of the suspension index and fracture density values ​​in the free point cluster based on the connection point set includes:

[0094] A local fitted surface is constructed based on the set of connection points, and then the surface area of ​​the local fitted surface is calculated as the connection area.

[0095] Analyze the minimum bounding geometry of each free point cluster in the free point cluster set, and calculate the suspended volume;

[0096] Divide the suspended volume by the connected area to obtain the suspension index;

[0097] Extract the neighborhood point cloud of the connection point set on one side of the main rock mass to obtain the point cloud of the parent rock in the connection zone;

[0098] The normal vector of each point in the point cloud of the parent rock in the connection zone is calculated. Then, the dispersion of each vector direction in the normal vector is statistically analyzed and quantified into a scalar value to obtain the fracture density value.

[0099] In this embodiment of the invention, the Ball Pivoting Algorithm is used to process the connection point set of 850 points obtained in the previous step. This algorithm constructs a triangular mesh surface model by simulating a virtual ball rolling on the surface of the point cloud. The radius of the virtual ball is set to 20 centimeters, which is slightly larger than the average spacing of the point cloud. This value is sufficient to cross tiny holes without excessively smoothing out important geometric features. Finally, a locally fitted surface composed of 1,580 tiny triangles is generated. By accumulating the areas of all triangles, the total surface area of ​​the surface is calculated to be 0.8 square meters, which is then determined as the connection area.

[0100] It should be noted that using minimum bounding geometry (especially minimum bounding boxes with variable orientation) can more compactly and accurately wrap irregularly shaped point clouds compared to axis-aligned bounding boxes, thus avoiding overestimation of volume.

[0101] In one implementation of this invention, for cluster 1 (containing 15,340 points) in the free point cluster set, principal component analysis (PCA) is applied. First, the three principal directions (eigenvectors) of the point cloud distribution are calculated. Then, the point cloud is rotated to align with these principal directions, and its maximum and minimum coordinate ranges in this new coordinate system are calculated, thereby constructing the minimum bounding box of the point cloud. Specifically, the calculated length, width, and height of this minimum bounding box are 2.0 meters, 1.0 meter, and 0.6 meters, respectively, therefore its volume is 2.0. 1.0 0.6 = 1.2 cubic meters, and this value is determined as the suspended volume.

[0102] Specifically, the suspension index is calculated using the following mathematical formula. :

[0103] ;

[0104] in: The suspension index (dimensionless) indicates that the larger the value, the larger the volume supported by a unit connection area, and the more unstable the structure. This represents the suspended volume calculated in the previous step (unit: cubic meters). The connected area (unit: square meters) obtained in the first step; substitute the obtained value into the formula: =1.2 cubic meters / 0.8 square meters = 1.5, therefore, the suspension index of this free point cluster is determined to be 1.5.

[0105] In one implementation of this invention, from a set of 850 connection points, all points belonging to the main rock mass (assuming there are 420 points) are selected. Then, a radius search is performed within the main rock mass point cloud centered on these 420 points. The search radius is set to 30 centimeters. This radius is intended to include enough environmental information near the connection points while avoiding the introduction of too far or irrelevant point cloud data. All main rock mass points within the search radius are merged, and duplicate points are removed. Finally, a subset of point clouds containing 2,500 points is obtained, which is defined as the parent rock point cloud of the connection zone.

[0106] In this embodiment of the invention, for each point in the point cloud of the parent rock in the connecting area, a local plane is fitted using its 20 nearest neighbors, and the normal vector of this plane is used as the normal vector of that point. After calculating the normal vectors of all 2,500 points, calculate the average vector of all normal vectors. Then calculate each normal vector With this average vector The angle between Finally, calculate all these included angles. Standard deviation And use it as the fracture density value Therefore, assuming the final calculated standard deviation of the angle is 0.4 radians, the fracture density value of the connecting region is determined to be 0.4. The larger this value is, the more discrete the normal vector distribution is, the more uneven the rock mass surface is, and the more developed the potential fractures are.

[0107] Preferably, step S3 involves determining whether a high-risk detour area exists based on the risk attribution area. If such an area exists, a risk detour path is dynamically planned to adjust the single pre-scan flight path, including:

[0108] If the risk-attributed area is a normal flight area, the drone will continue to perform a single pre-scan flight.

[0109] If the risk-attributed area is a high-risk detour area, then the spatial impact radius is set according to the high-risk detour area;

[0110] Centered on the centroid of the high-risk detour area, a Gaussian potential field function is constructed within the spatial influence radius based on the potential energy threat value of the dangerous rock and the structural stability index, which reduces the risk intensity with distance. The high-risk detour area is then extended outward by a safe distance to obtain flight risk potential field data. Specifically, the potential energy threat value of the dangerous rock is mapped to the peak intensity of the Gaussian potential field function, and the reciprocal of the structural stability index is used as a parameter affecting its decay rate.

[0111] Dynamically plan risk detour routes based on flight risk potential field data;

[0112] By dynamically adjusting and replanning the single pre-scan route through risk detour paths, intelligent and safe execution routes are generated.

[0113] In this embodiment of the invention, all identified clusters of free points were analyzed. If the calculated structural stability index was higher than the preset high-risk screening threshold (e.g., 1.5), then the risk-assigned areas of the current exploration area were determined to be normal flight areas. Therefore, the flight control system will not initiate the path replanning procedure, but will directly execute the original single pre-scan flight path set by sonar detection in step S1, that is, instruct the UAV to fly in a straight line along the predetermined 33.8-meter central axis to the destination.

[0114] In another implementation of this invention, it is assumed that the free point cluster 1 identified in step S2 is determined to be a high-risk free point cluster, thus confirming the existence of a high-risk detour area. Specifically, the minimum bounding box of the high-risk free point cluster 1 is calculated, and its diagonal length is found to be 2.3 meters. Then, a safety buffer distance of 5 meters is added to this to set the spatial influence radius. Therefore, the final determined spatial influence radius is 2.3 meters + 5 meters = 7.3 meters, which means that the subsequent risk potential field will be effective within this radius.

[0115] In this embodiment of the invention, the centroid coordinates of the high-risk free point cluster 1 are calculated. Then construct a three-dimensional Gaussian potential function. Its mathematical expression is:

[0116] ;

[0117] in: For any point in space The risk intensity value; The peak intensity of the Gaussian potential field is given by the threat value of the boulders' potential energy. Mapped from, , σ is a preset weighting coefficient used to scale the potential threat value to an appropriate level of risk cost; here it is set to 10. σ is the standard deviation of the Gaussian function, which determines the decay rate (range of influence) of the potential field, and is determined by the reciprocal of the structural stability index (SSI). , It is a preset scaling factor used to adjust the size of the influence range; here it is set to 2.

[0118] In this embodiment of the invention, a fast extended random tree star algorithm is used to perform path search in the flight risk potential field data, with the starting point (current position of the UAV) of the original single pre-scan flight path as the starting point and the ending point (33.8 meters) as the target point.

[0119] In one implementation of this invention, the entry and exit points where the original single pre-scan flight path (38-meter straight line) intersects with the flight risk potential field data are determined. The straight line segment between the entry and exit points in the original flight path is deleted. Then, the risk detour path consisting of a series of waypoints planned in the previous step is inserted into the deleted position.

[0120] In another implementation of this invention, to ensure the smoothness of the UAV's flight, the spliced ​​waypoint sequence is smoothed using B-spline curves to generate a continuous flight path with varying curvature. This new, safe, and feasible path is then encapsulated and named the intelligent safe execution path, and subsequently sent to the UAV's flight control system for execution.

[0121] Preferably, dynamically planning risk detour routes based on flight risk potential field data includes:

[0122] The starting and target points of the detour path are determined from the flight risk potential field data based on a single pre-scan flight path.

[0123] Define a path cost function, which includes three components: path length, total cumulative risk potential field value traversed, and flight energy consumption.

[0124] A fast extended random tree search is performed on the flight risk potential field data using the path cost function to explore feasible paths connecting the starting point and the target point;

[0125] When the convergence condition is met or the maximum number of iterations is reached, output the feasible path with the lowest current cost as a risk avoidance path.

[0126] In one implementation of this invention, the target point of the detour path is set as the endpoint coordinate of the original single pre-scan flight path. This coordinate is calculated by translating the starting point coordinate by 33.8 meters along the positive direction vector of the UAV's nose. These two points together define the solution boundary of the path planning problem.

[0127] In one implementation of this invention, the path cost function is set. The specific mathematical expression is:

[0128] ;

[0129] in: This represents a candidate path from the starting point to the target point. For path The total path length is obtained directly by integrating the path. For path The total cumulative risk potential value of the path is obtained by summing the risk intensity values ​​of each grid cell traversed by the path. For path The calculation model for estimating flight energy consumption is as follows: ,in It is the curvature of the path. It is the change in total elevation. These are energy consumption coefficients, representing energy consumption related to distance, turning, and climb, respectively. , , These are the weighting coefficients for the three components, set according to task priority, and their sum is 1. In this case, to prioritize safety, the three are set to 0.2, 0.6 and 0.2 respectively.

[0130] In one implementation of this invention, the algorithm starts from the starting point, randomly generates sampling points in three-dimensional space, finds the node closest to the sampling point in the tree, and extends it in the direction of the sampling point by a fixed step size (e.g., 1 meter). If the newly generated node and its connecting path do not collide with any area where the risk value exceeds a preset safety threshold (e.g., 80% of the peak intensity), the new node is added to the tree, and its neighboring nodes are checked. The algorithm attempts to reconnect through the new node to optimize the path cost. This process is repeated until a branch of the tree reaches the vicinity of the target point.

[0131] In this embodiment of the invention, this step defines the termination conditions of the search algorithm and outputs the final planning result when the conditions are met. Specifically, two parallel termination conditions are set: one is the convergence condition, that is, in 100 consecutive iterations, the improvement in the cost value of the optimal path found is less than 0.1%, which indicates that the algorithm has basically converged; the other is the maximum number of iterations, that is, the total number of iterations executed by the algorithm reaches a preset upper limit of 5000 times, to ensure that the planning process is completed within a limited time.

[0132] In one implementation of this invention, it is assumed that a feasible path connecting the starting point and the target point is found on the 3800th iteration, and its cost function value decreases from 125.8 to 125.7 in the next 100 iterations, with an improvement rate of less than 0.1%, satisfying the convergence condition. Therefore, the search process is immediately terminated, and the feasible path with a cost value of 125.7 recorded in the current tree is output as the final planning result. This path is thus identified as the risk detour path.

[0133] Of particular importance is that when the area ahead is identified as a high-risk detour area, the route planning process specifically includes:

[0134] Extract high-risk free point clusters within the high-risk detour area and lock their geometric center points as the target that the sensor is pointing at, defining them as the target to be investigated;

[0135] The command drone immediately stops moving forward and adjusts its position to maintain a preset safe observation distance from the target to be surveyed, so as to enter the survey standby state;

[0136] Based on intelligent and safe execution of route planning, survey and scanning waypoints are covered from multiple key observation perspectives;

[0137] Based on the reconnaissance standby status, the gimbal pitch and yaw angle commands are set for the reconnaissance scanning waypoints to ensure that the laser scanner is always focused on the target to be reconnaissance throughout the flight, so as to obtain a multi-view waypoint series.

[0138] By dynamically adjusting the scanning line density of the laser scanner, localized intensified scanning of multi-view waypoint columns is performed to obtain enhanced survey data.

[0139] In one implementation of this invention, the high-risk free point cluster (i.e., cluster 1 identified in the previous step) that triggers the generation of the high-risk detour area is directly extracted from the definition data of the high-risk detour area. The three-dimensional coordinates of the 15,340 points contained in cluster 1 are traversed. The geometric center point coordinates of the point cluster are calculated by averaging the X, Y, and Z coordinates of all points, assumed to be (15.2, 8.5, 9.1). The coordinate point is locked and formally defined as the target to be explored in the task.

[0140] In one implementation of this invention, the preset safe observation distance is set to 8 meters. This distance is determined by comprehensively considering the optimal working range of the laser scanner and the sufficient reaction space reserved for emergencies. The direction vector between the current position of the UAV and the target to be surveyed is calculated, and the UAV is moved to a position point exactly 8 meters away from the target along the opposite direction of this vector. After the UAV hovers stably at this position, its status is updated to the survey standby state.

[0141] In this embodiment of the invention, with the target to be surveyed as the center and a safe observation distance of 8 meters as the radius, seven survey scanning waypoints are planned on a safe hemisphere tangent to the original intelligent safe execution route. These waypoints are distributed in an arc shape, covering the key observation angle from 45 degrees to the left and right front of the target to be surveyed. Specifically, the seven survey scanning waypoints generated in the previous step are traversed, and for each waypoint... Calculate the distance from this waypoint to the target to be surveyed. Direction vector .

[0142] In one implementation of this invention, the direction vector is utilized. To calculate the gimbal yaw angle that the drone needs to be set at this waypoint. and pitch angle :

[0143] ;

[0144] ;

[0145] in: It is a two-parameter arctangent function used for accurate calculation of yaw angle. It is a vector The length of the module.

[0146] In this embodiment of the invention, before the UAV begins its multi-view waypoint flight mission, a command is sent to the controller of the laser scanner to dynamically reduce its rotational scanning frequency from 20 Hz for navigation to 5 Hz for fine scanning. It should be noted that a lower scanning frequency means that the distribution of the laser scanning beam on the target surface becomes denser per unit time, thus significantly increasing the density of the scanned point cloud. Therefore, after the UAV flies along the multi-view waypoint and continuously scans the target, the point cloud data acquired has approximately four times higher point density on the target surface compared to conventional scanning. This locally encrypted high-density point cloud set is defined as enhanced survey data.

[0147] The present invention also provides a laser scanning-based UAV path planning system, which executes the laser scanning-based UAV path planning method described above. The laser scanning-based UAV path planning system includes:

[0148] The flight range scanning perception module S101 is used to set a single pre-scan flight range, instruct the UAV to fly along the flight range, and use a laser scanner to scan the tunnel environment in real time to obtain the original radial profile point cloud data.

[0149] The real-time risk assessment module S102 is used to separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path;

[0150] The intelligent route generation module S103 is used to determine whether there is a high-risk detour area based on the risk attribution area. If there is, it dynamically plans the risk detour route to adjust the single pre-scan route and generate an intelligent and safe execution route.

[0151] The navigation control and iteration module S104 is used to instruct the UAV to fly along the intelligent and safe execution route to its destination, and to use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

[0152] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0153] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement 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 of the invention herein.

Claims

1. A path planning method for unmanned aerial vehicles (UAVs) based on laser scanning, characterized in that, Includes the following steps: Step S1: Set a single pre-scan flight path, instruct the UAV to fly along the flight path, and use a laser scanner to scan the tunnel environment in real time to obtain the original radial profile point cloud data; Step S2: Separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path; before separating the free point clusters based on the original radial profile point cloud data, the following steps are also included: The original radial profile point cloud data is rasterized in three dimensions to obtain three-dimensional scan point cloud data. The 3D scan point cloud data is divided into spatial voxel grids, and then the laser pulse points in the 3D scan point cloud data are assigned to the corresponding spatial voxel grids to obtain voxelized pulse data. Based on the voxelized pulse data, the Euclidean distance between the three-dimensional coordinate geometric center of all laser pulse points in each voxel and the corresponding voxel geometric center is calculated to obtain the positional eccentricity. Simultaneously, the standard deviation of the echo intensity values ​​of all laser pulse points within each voxel in the voxelized pulse data is calculated to obtain the intensity fluctuation value. Determine whether the positional eccentricity is less than the preset geometric aggregation threshold and whether the intensity fluctuation value is less than the preset material uniformity threshold. If so, mark the voxel as a high-confidence voxel. Based on high-confidence voxel analysis of the centroids of pulse points within voxels, the set of all centroids is used to construct a structural confidence point cloud. Among them, the free point clusters separated based on the original radial profile point cloud data include: Perform a region growing operation on the structural credibility point cloud to identify and extract the largest connected point cloud set, and determine this set as the main rock mass; Perform Boolean difference operations between the structural credibility point cloud and the main rock mass to separate all points that do not belong to the main rock mass, thus obtaining the initial suspended body; Cluster analysis using Euclidean distance is performed on the initial suspended body to segment the scattered point cloud on the initial suspended body into multiple independent point cloud clusters, thus obtaining a set of point cloud clusters; Calculate the number of points in each cluster of the point cloud cluster set, and filter out the tiny clusters whose number is lower than the preset noise point threshold. The remaining cluster set is the free point cluster set. Step S3: Determine whether there is a high-risk detour area based on the risk attribution area. If so, dynamically plan the risk detour route to adjust the single pre-scan flight path and generate an intelligent and safe execution route. Step S4: Instruct the UAV to fly along the intelligent and safe execution route to its destination, and use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

2. The UAV path planning method based on laser scanning according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Before the UAV enters the target underground mine space, use its onboard broadband sonar detector to perform fan-shaped acoustic detection on the area in front to obtain initial acoustic echo data containing the main contour reflection information. Step S12: Based on the farthest echo boundary identified in the initial acoustic echo data, calculate and set a safe forward distance to obtain the single pre-scan range; Step S13: Instruct the UAV to fly along the central axis of the single pre-scan flight path to the end of the single pre-scan flight path. During this process, use the dual-axis rotating laser scanner that is orthogonal to the flight direction to perform real-time radial scanning of the roadway sides and top and bottom plates of the target underground mine space to obtain the original radial profile point cloud data.

3. The UAV path planning method based on laser scanning according to claim 1, characterized in that, Step S2 analyzes the structural stability index and potential energy threat value of the free point cluster to determine the risk attribution area of ​​the current UAV flight path, including: Search for the set of points whose spatial distance from the cluster of free points to the surface points in the main rock mass is less than a preset connection distance threshold, and define this set of points as the connection point set. Based on the analysis of the connection point set, the suspension index and fracture density value of the free point cluster are analyzed; The structural stability index is obtained by weighting the suspension index and the crack density value to evaluate the structural stability of the connection area. High-risk free point clusters in the free point cluster set are selected based on preset high-risk screening thresholds and structural stability index. Calculate the local curvature or surface variation of each point in the high-risk free point cluster, and identify the point set with curvature or surface variation greater than the preset convexity threshold as a convex shape, so as to segment out isolated convex clusters; Assess the potential threat value of dangerous rocks based on isolated clusters of protrusions; The risk attribution area of ​​the current UAV flight path is determined based on the structural stability index and the potential energy threat value of the dangerous rock; the risk attribution area includes the normal flight area and the high-risk detour area.

4. The UAV path planning method based on laser scanning according to claim 3, characterized in that, The assessment of the potential threat value of unstable rocks based on isolated clusters of protrusions includes: Construct the minimum convex hull of all outer points in an isolated cluster of protrusions to obtain the dangerous rock protrusion body; The number of all non-empty voxel units containing at least one point cloud data within the envelope of an isolated convex cluster is counted, and this number is multiplied by the volume of a single voxel unit to obtain the solid filling product. The solid filling volume is compared with the total volume of the unstable rock mass, and the ratio of the degree of development of internal cavities or fissures is calculated as the structural porosity. The potential energy threat value of the unstable rock mass is determined based on its volume and structural looseness, and according to the height of the centroid of the isolated cluster of protrusions relative to the roadway floor.

5. The UAV path planning method based on laser scanning according to claim 3, characterized in that, The analysis of the suspension index and fracture density values ​​in the free point cluster based on the connection point set includes: A local fitted surface is constructed based on the set of connection points, and then the surface area of ​​the local fitted surface is calculated as the connection area. Analyze the minimum bounding geometry of each free point cluster in the free point cluster set, and calculate the suspended volume; Divide the suspended volume by the connected area to obtain the suspension index; Extract the neighborhood point cloud of the connection point set on one side of the main rock mass to obtain the point cloud of the parent rock in the connection zone; The normal vector of each point in the point cloud of the parent rock in the connection zone is calculated. Then, the dispersion of each vector direction in the normal vector is statistically analyzed and quantified into a scalar value to obtain the fracture density value.

6. The UAV path planning method based on laser scanning according to claim 1, characterized in that, Step S3 involves determining whether a high-risk detour area exists based on the risk attribution area. If such an area exists, a risk detour path is dynamically planned to adjust the single pre-scan flight path, including: If the risk-attributed area is a normal flight area, the drone will continue to perform a single pre-scan flight. If the risk-attributed area is a high-risk detour area, then the spatial impact radius is set according to the high-risk detour area; Centered on the centroid of the high-risk detour area, a Gaussian potential field function is constructed within the spatial influence radius based on the potential energy threat value of the dangerous rock and the structural stability index, which reduces the risk intensity with distance. The high-risk detour area is then extended outward by a safe distance to obtain flight risk potential field data. Specifically, the potential energy threat value of the dangerous rock is mapped to the peak intensity of the Gaussian potential field function, and the reciprocal of the structural stability index is used as a parameter affecting its decay rate. Dynamically plan risk detour routes based on flight risk potential field data; By dynamically adjusting and replanning the single pre-scan route through risk detour paths, intelligent and safe execution routes are generated.

7. The UAV path planning method based on laser scanning according to claim 6, characterized in that, Dynamically planning risk detour routes based on flight risk potential field data includes: The starting and target points of the detour path are determined from the flight risk potential field data based on a single pre-scan flight path. Define a path cost function, which includes three components: path length, total cumulative risk potential field value traversed, and flight energy consumption. A fast extended random tree search is performed on the flight risk potential field data using the path cost function to explore feasible paths connecting the starting point and the target point; When the convergence condition is met or the maximum number of iterations is reached, output the feasible path with the lowest current cost as a risk avoidance path.

8. A UAV path planning system based on laser scanning, characterized in that, For executing the laser scanning-based UAV path planning method as described in claim 1, the laser scanning-based UAV path planning system includes: The flight range scanning perception module is used to set a single pre-scan flight range, instruct the UAV to fly along the flight range, and use a laser scanner to scan the tunnel environment in real time to obtain raw radial profile point cloud data. The real-time risk assessment module is used to separate free point clusters based on the original radial profile point cloud data; analyze the structural stability index and potential energy threat value of the free point clusters to determine the risk attribution area of ​​the current UAV flight path; The intelligent route generation module is used to determine whether there are high-risk detour areas based on the risk attribution area. If so, it dynamically plans risk detour routes to adjust the single pre-scan route and generate intelligent and safe execution routes. The navigation control and iteration module is used to instruct the UAV to fly along a smart and safe execution route to its destination, and use the destination as the starting point for the next round of path planning, until the iterative exploration of the target underground mine space is completed.

Images