Search

CN-121740061-B - Expressway unmanned aerial vehicle autonomous route planning method and system based on A-algorithm

CN121740061BCN 121740061 BCN121740061 BCN 121740061BCN-121740061-B

Abstract

The invention belongs to the technical field of unmanned aerial vehicle autonomous navigation and intelligent inspection, and particularly relates to a highway unmanned aerial vehicle autonomous route planning method and system based on an A-algorithm, wherein the method comprises the steps of obtaining three-dimensional point cloud data of a highway, loading, downsampling, filtering, classifying and extracting a central line of the three-dimensional point cloud data; the method comprises the steps of modeling based on point cloud data, specifically comprising voxel grid map construction and obstacle attribute embedding, planning an autonomous route of the unmanned aerial vehicle by adopting an A-type algorithm, and smoothing an initial path. The invention realizes double breakthrough of precision and efficiency on technical performance, reduces cost and increases efficiency on economic cost, and realizes full scene risk prevention and control on safety guarantee.

Inventors

  • SUN HESHAN
  • DAI JUN
  • ZHANG YU
  • ZHOU XIAO
  • WANG YANG

Assignees

  • 辽宁盘古技术有限公司

Dates

Publication Date
20260512
Application Date
20260228

Claims (10)

  1. 1. The method for planning the autonomous route of the expressway unmanned aerial vehicle based on the A-algorithm is characterized by comprising the following steps of: step 1, three-dimensional point cloud data of a highway are obtained, and the three-dimensional point cloud data are loaded, downsampled, filtered, classified and extracted by a central line; And 2, modeling based on the point cloud data processed in the step 1, wherein the modeling comprises the following steps: step 201, constructing a voxel grid map, namely dividing a highway inspection area into three-dimensional voxel grids, wherein an X axis and a Y axis correspond to plane positions, and a Z axis corresponds to height information; Step 202, embedding barrier attributes, namely in a constructed voxel grid map, carrying out spatial indexing on classified point cloud data according to voxel coordinates, and embedding multidimensional attribute information into each voxel unit to form a voxel node library to be explored; and 3, planning an autonomous route of the unmanned aerial vehicle by adopting an A-algorithm, wherein the method specifically comprises the following steps of: step 301, converting the global coordinate system into a local coordinate system; Step 302, designing a cost function as follows: F(n) = G(n) + H(n) +αWm +βWs +γWo; Wherein G (n) is an actual cost from the starting point to the current voxel node n, and includes a distance cost L (n), a gradient cost P (n), and a meteorological cost W (n), which are specifically expressed as follows: G(n) =ω 1 L(n) +ω 2 P(n) +ω 3 W(n); Wherein, the P (n) =k_p·|arctan ((z n+1 -z n )/l_xy) |, k_p is a gradient coefficient, l_xy is a straight line distance on the X-Y plane of the current voxel node n and the next voxel node n+1, i.e. "horizontal distance", W (n) =k_w·max (0, v_wind-v_safe), k_w is a meteorological coefficient, v_wind is a wind speed penalty term, v_safe is a wind speed penalty term, ω 1 、ω 2 、ω 3 is an actual cost sub-term weight coefficient; h (n) is a heuristic cost from the current voxel node n to the target voxel node, and is estimated by using a three-dimensional euclidean distance: ; wm, ws and Wo are weather, gradient and obstacle penalty items respectively, and alpha, beta and gamma are penalty item weights; step 303, selecting the optimal voxel node with the minimum cost from the voxel node library to be explored in step 202 according to the cost function of step 302, gradually searching the target voxel node from the voxel nodes of the initial path to form the initial path; And 4, performing post-processing on the initial path planned in the step 3, wherein the post-processing specifically comprises the following steps: Step 401, smoothing the initial path by adopting a 3-order B spline curve, setting the interval of voxel nodes to be 5m, and for a control point sequence P 0 ,P 1 ,...,P n , representing any point on the curve as follows: P(u) =Σ k =0 3 P i+k ·N k,3 (u), u∈[0,1]; Wherein the basis functions: N 0,3 (u) = (1-u) 3 /6; N 1,3 (u) = (3u 3 -6u 2 +4)/6; N 2,3 (u) = (-3u 3 +3u 2 +3u+1)/6; N 3,3 (u) = u 3 /6; Wherein, P i+k is an "original reference point" for generating a B-spline curve, i.e. an "initial path voxel node" planned in step 3, and since the 3-order B-spline curve satisfies C2 continuity at the connection point, curvature κ= |p ' (u) ×p ' (u) |/|p ' (u) | 3 is continuous, satisfying the unmanned power constraint; step 402, designing dynamic height obstacle avoidance logic, and automatically adjusting safety margin according to the type of the obstacle.
  2. 2. The method for autonomous highway unmanned aerial vehicle planning based on the a-algorithm according to claim 1, wherein the step1 comprises the steps of: step 101, loading three-dimensional point cloud data to obtain LAS-format point cloud data; Step 102, downsampling by adopting a voxel grid downsampling method, setting the voxel size to be 0.5 m, and reducing the point cloud data density to 50 points/m 2 ; step 103, noise removal is carried out by using a statistical filtering algorithm, a threshold k=10 of the number of neighborhood points is set, the standard deviation multiple is 2, and points with the average distance exceeding 2 times of the standard deviation with the neighborhood points are judged as noise points and removed; step 104, classifying the filtered point cloud data based on geometric features by utilizing CANUPO plug-in units to distinguish different types of targets; and 105, extracting the center line of the expressway based on the center line extraction method of the principal component analysis.
  3. 3. The method for autonomous highway unmanned aerial vehicle planning based on the algorithm of claim 2, wherein in step 104, the different types of targets include ground, road markings, guardrails, trees, utility poles, high voltage lines, traffic signs, street lamps, bridge piers, tunnel entrance/exit structures, emergency kiosks, construction area obstacles.
  4. 4. The autonomous highway unmanned aerial vehicle planning method according to claim 1, wherein in step 201, the grid size is set to be 1m x 0.5m.
  5. 5. The autonomous highway unmanned aerial vehicle planning method based on the a-algorithm according to claim 1, wherein in the step 202, the multidimensional attribute information comprises an obstacle type, an altitude, a threat level and a dynamic update time stamp, the method is characterized in that the method is used for marking a no-fly zone as non-passable, and the method is used for assigning corresponding weight values to weather sensitive zones according to weather data, wherein the weight values are dynamically adjusted along with the change of weather conditions.
  6. 6. The autonomous highway unmanned aerial vehicle planning method according to claim 1, wherein in step 202, for the elongated obstacle in the highway scene, a fine modeling algorithm based on skeleton extraction is designed, comprising the steps of: Step 202-1, target segmentation, namely screening a point cloud subset of the elongated obstacle from the classified point clouds; step 202-2, skeleton extraction, namely extracting a central skeleton line of an obstacle from a point cloud subset by adopting a distance transformation plus Zhang-Suen refinement algorithm; Step 202-3, voxel correction, namely changing an original single voxel mark into a plurality of small voxel clusters along a skeleton line along a central skeleton line, and embedding an elongated attribute; And 202-4, expanding a collision radius, namely additionally marking a 'safe buffer zone voxel' around a skeleton voxel, and adapting to the obstacle avoidance requirement of the unmanned aerial vehicle.
  7. 7. The autonomous highway unmanned aerial vehicle planning method according to claim 1, wherein in step 301, the global coordinate system is a universal geographic coordinate system, the coordinates are (x_global, y_global, z_global) for locating absolute spatial positions, the local coordinate system is a "path coordinate system" based on the highway centerline, defined as (S, D, Z), the S-axis is a mileage distance along the centerline, i.e., a centerline length from the starting point to the current point, the D-axis is a lateral offset, i.e., a left-right distance with respect to the centerline, D >0 is a right side, D <0 is a left side, the Z-axis is altitude information, and the global coordinate system is consistent with the global coordinate system Z-axis, and the converting the global coordinate system into the local coordinate system comprises the steps of: step 301-1 of determining origin and pose of local coordinate System based on centerline Selecting a central line point of a highway starting point in a global coordinate system as a local origin O' (X0, Y0, Z0), defining a local coordinate system S axis along the tangential direction of the central line, defining a D axis along the horizontal direction of a vertical central line, and defining a Z axis along the vertical direction to form a posture matrix R of the local coordinate system as a rotation matrix; step 301-2 mathematical transformation of global coordinates to local coordinates Translation correction, namely subtracting the local origin coordinates from the global coordinates, and eliminating origin differences: ; rotation alignment by rotating the matrix R to project the translated coordinates to the (S, D) plane of the local coordinate system, with the Z axis unchanged: ; The parameters of the rotation matrix R are calculated from the real-time tangential direction of the central line, so that the S-axis is ensured to extend along the central line of the current road section all the time; And the parameter mapping is that the S value is obtained by calculating the arc length of the center line from the point to the origin after translational rotation, and the D value is the vertical distance from the point to the center line.
  8. 8. The method for autonomous highway unmanned aerial vehicle planning based on the a-algorithm according to claim 1, wherein the step 303 comprises the following specific steps: Step 303-1, initializing, namely adding an initial path voxel node into a voxel node library to be explored, calculating F (n) of the initial path voxel node, and then moving the initial path voxel node into the explored voxel node library; selecting a voxel node with the minimum F (n) from a voxel node library to be explored as a current voxel node, and moving to the explored voxel node library; Step 303-3, neighborhood searching, namely, searching 26 adjacent voxel nodes of the current voxel node through a 26 neighborhood searching strategy, and skipping the voxel node marked as 'unpunctable'; step 303-4, calculating cost, namely calculating F (n) for each feasible adjacent voxel node; Step 303-5, updating the voxel node, namely if the adjacent voxel node is not in the voxel node library to be explored or F (n) of the adjacent voxel node is smaller, updating a parent node of the adjacent voxel node as a current voxel node, and adding the explored voxel node library; And step 303-6, terminating, namely backtracking all father nodes from the target voxel nodes to the initial path voxel nodes when the target voxel nodes are added into the explored voxel node library, and obtaining a voxel node sequence which is the initial path.
  9. 9. The method for autonomous lane planning of a highway unmanned aerial vehicle based on an a-algorithm according to claim 1, wherein in the step 402, a 3m vertical margin is adopted for a flexible obstacle, a 5m vertical margin is adopted for a rigid obstacle, a 10m vertical margin is adopted for a large-scale structure, and on the basis of a waypoint generation strategy, the waypoint density is dynamically adjusted according to a task type, namely a variable waypoint interval of 20-200 m is set for a daily routing inspection task, and an encrypted waypoint of 20-50 m is adopted for an emergency task.
  10. 10. The system for planning the autonomous air route of the expressway unmanned aerial vehicle based on the A-algorithm is characterized by comprising the following modules based on the planning method of claim 1: The data processing module is used for acquiring three-dimensional point cloud data of the expressway, and carrying out loading, downsampling, filtering, classification and central line extraction on the three-dimensional point cloud data; the three-dimensional modeling module is used for modeling based on the point cloud data processed by the data processing module; The path planning module is used for planning an autonomous route of the unmanned aerial vehicle by adopting an A-type algorithm; and the path optimization module is used for carrying out smoothing treatment and dynamic obstacle avoidance on the initial path generated by the path planning module.

Description

Expressway unmanned aerial vehicle autonomous route planning method and system based on A-algorithm Technical Field The invention belongs to the technical field of unmanned aerial vehicle autonomous navigation and intelligent inspection, and particularly relates to a highway unmanned aerial vehicle autonomous route planning method and system based on an A-algorithm. Background Expressway inspection is used as a key link for guaranteeing the safe operation of road infrastructure, and the traditional inspection mode faces double challenges of efficiency and safety. At present, two modes of manual inspection and fixed equipment monitoring are mainly relied on in the industry, wherein manual inspection requires workers to drive vehicles or hiking inspection, efficiency bottleneck that inspection takes more than 2 hours per hundred kilometers exists, traffic accident risks in a high-speed environment are faced, all-weather monitoring can be realized through fixed equipment monitoring, coverage dead areas exist, high cost is additionally increased for upgrading high-precision equipment such as laser radars, and the burden of small and medium operation units is difficult. The application of the existing unmanned aerial vehicle path planning technology in expressway scenes still has significant limitations. In the space modeling aspect, a main stream method adopts two-dimensional plane abstraction, three-dimensional terrain features such as bridge gradient, tunnel height difference and the like cannot be accurately represented, so that the unmanned aerial vehicle is prone to collision risk on a complex road section, a traditional global coordinate system lacks dynamic adaptation capability in terms of coordinate system selection, when the unmanned aerial vehicle encounters strong crosswind and other interferences, path correction response delay exceeds 3 seconds, a rule engine design mostly adopts external judgment logic, constraint conditions such as a limited flight zone, a forbidden flight height and the like are separated from a path search algorithm, planning efficiency is reduced by 27%, and in terms of path smoothness, a large number of broken line inflection points exist based on a planning result of a grid method, the gesture is required to be frequently adjusted in the execution process of the unmanned aerial vehicle, and battery endurance time is shortened by 15% -20%. From the actual demands of the industry, the large-scale application of unmanned aerial vehicle inspection is limited by two core pain points, namely that the number of flying hands with civil aviation certification qualification is more than 1.2 ten thousand people, the manual operation cost accounts for 63% of the total inspection cost, and the response capability of the existing system in an emergency scene is insufficient, after sudden accidents (such as vehicle rear-end collision and road surface collapse) occur, the unmanned aerial vehicle takes 14 minutes on average from taking off to building a stable monitoring route, and the real-time data demand of a gold rescue period is difficult to meet. These technical drawbacks, together with the industry pain, highlight the necessity and urgency of developing a new way of autonomous lane planning for a highway unmanned aerial vehicle. The limitation of the conventional technical path has become a key factor for restricting the deep application of unmanned aerial vehicles in the intelligent traffic field. Along with the annual increase of expressway mileage by 6.5 percent and the increase of bridge tunnel ratio to 22 percent, an integrated solution capable of integrating three-dimensional space perception, dynamic constraint decision and intelligent path optimization is needed to be constructed so as to break through the bottleneck of the prior art and promote the expressway inspection to unmanned and intelligent upgrading. Disclosure of Invention In order to solve the technical problems, the invention provides a highway unmanned aerial vehicle autonomous route planning method and system based on an A-algorithm. The invention provides an automatic route planning method of a highway unmanned aerial vehicle based on an A algorithm, which comprises the following steps: step 1, three-dimensional point cloud data of a highway are obtained, and the three-dimensional point cloud data are loaded, downsampled, filtered, classified and extracted by a central line; And 2, modeling based on the point cloud data processed in the step 1, wherein the modeling comprises the following steps: step 201, constructing a voxel grid map, namely dividing a highway inspection area into three-dimensional voxel grids, wherein an X axis and a Y axis correspond to plane positions, and a Z axis corresponds to height information; Step 202, embedding barrier attributes, namely in a constructed voxel grid map, carrying out spatial indexing on classified point cloud data according to voxel coordinates, and embedding multidi