Unmanned aerial vehicle mixed path planning method based on track inspection and storage medium

Abstract

The invention provides an unmanned aerial vehicle mixed path planning method based on track inspection and a storage medium, and the method comprises the steps: planning a preliminary global inspection path through an A * algorithm, dividing the global path into a plurality of local inspection paths, and recognizing whether there is a random obstacle in each local inspection path, and re-planning a local inspection path with a random obstacle by using a DQN algorithm. The invention further comprises a method for identifying and avoiding dynamic obstacles in each section of local inspection path. According to the method, the advantages of the A * algorithm and the DQN algorithm are fused, when a random obstacle or a dynamic obstacle appears in a certain section of local inspection path, only the section of local inspection path needs to be adjusted, the global inspection path does not need to be planned again, the unnecessary calculation process is reduced, and the path planning efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) path planning, and in particular to a hybrid UAV path planning method based on track inspection. Background Art

[0002] Most of the existing track inspections are completed by inspection robots. Since the inspection robots rely on preset inspection routes, they lack real-time data processing capabilities and cannot provide timely feedback on the on-site situation. Therefore, drones are gradually being used in track inspections. Since drones fly in the air, their visual range is limited and it is impossible to set a complete inspection route in advance. In addition, there may be moving obstacles on the railway tracks, so drones need to have better obstacle avoidance capabilities and adjust the inspection path in time. Current drones do not have this function, so there is an urgent need to improve the path planning of existing drones used for track inspections. Summary of the invention

[0003] The present invention provides a hybrid path planning method for an unmanned aerial vehicle based on track inspection, which can solve the problem in the prior art that the unmanned aerial vehicle for track inspection cannot accurately avoid obstacles.

[0004] The objective of the present invention is achieved through the following technical solutions: The hybrid path planning method of UAV based on track inspection includes the following steps: Step S1, using the A* algorithm to plan a preliminary global inspection path; Step S2, extracting key points of the global inspection path, and dividing the global inspection path into N local inspection paths according to the key points; Step S3, the UAV flies along each local inspection path in sequence, monitors the local inspection path where it is located, and determines whether there are random obstacles in the local inspection path where it is located; if not, continue to fly until the local target point of the local inspection path, start monitoring the next local inspection path again, repeat step S3, until it reaches the end of the global inspection path, and end the inspection task; if yes, go to step S4; Step S4: when a random obstacle is detected in a local inspection path, the DQN algorithm is used to replan the local inspection path; Step S5: Whether the end point of the global inspection path is reached, if yes, the inspection task is terminated, otherwise, go to step S3.

[0005] Furthermore, it also includes methods for avoiding dynamic obstacles: The drone reads the image frame data of the depth camera in real time, processes the image frame data using the YOLOv5 target detection algorithm, identifies dynamic obstacles, and quickly and accurately locates and classifies dynamic obstacles; Use the DBSCAN algorithm to process the point cloud data generated by the depth camera and group the points belonging to the same dynamic obstacle together to form a cluster; Based on the clustering results of the DBSCAN algorithm and the dynamic obstacles identified by the YOLOv5 target detection algorithm, the points in the point cloud data that belong to the same dynamic obstacle are characterized, and the point cloud data that does not belong to the dynamic obstacle is separated; Calculate and analyze the type, distance and movement direction of dynamic obstacles, estimate the speed of dynamic obstacles, analyze the collision probability, and dynamically adjust the flight speed and direction of the drone to achieve effective avoidance of dynamic obstacles.

[0006] Furthermore, the specific process of replanning the local inspection path using the DQN algorithm is as follows: Take environmental state data as input features; The DQN algorithm outputs action instructions by identifying environmental state data; The drone executes the action instructions output by the DQN algorithm.

[0007] Furthermore, the environmental status data includes the shape, size and position of random obstacles, and the position of the drone.

[0008] Furthermore, the action instructions include but are not limited to continue moving forward, turning left, turning right or retreating.

[0009] Furthermore, the DBSCAN algorithm is used to group points belonging to the same dynamic obstacle together to form a cluster, including: the DBSCAN algorithm calculates the density relationship between points in the point cloud data, and for a given point set, with a certain radius and a minimum number of included points as conditions, classifies closely adjacent points into one category, and groups points belonging to the same dynamic obstacle together to form a cluster.

[0010] The present invention also provides a storage medium on which a program is installed, characterized in that the above-mentioned hybrid path planning method for unmanned aerial vehicles based on track inspection can be executed by running the program.

[0011] The hybrid path planning method for UAV based on track inspection of the present invention avoids random obstacles that appear in the range and plans the inspection path within the range of vision for the extension of the UAV during flight. It can also reasonably avoid dynamic obstacles that suddenly appear in the planned inspection path and adjust the inspection path in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of a hybrid path planning method for a UAV according to Embodiment 1 of the present invention; Figure 2A schematic diagram for explaining the global path and local path of the present invention; Figures 3 to 6 This is a schematic diagram of a path planning process according to the first embodiment of the present invention; Figure 7 This is a flow chart of the dynamic obstacle recognition and avoidance of the present invention. DETAILED DESCRIPTION

[0013] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0014] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0015] Embodiment 1 The hybrid path planning method of the UAV based on track inspection in this embodiment includes the following steps: Step S1: Plan a preliminary global inspection path using the A* algorithm.

[0016] Before the drone performs a mission, it uses the A* algorithm to perform global path planning on the environment and determine the preliminary path from the inspection starting point to the inspection end point. The algorithm takes into account the static characteristics of the environment and the layout of obstacles to generate a path with the lowest cost.

[0017] The A* algorithm is a classic heuristic graph search algorithm and the most commonly used algorithm in global static path planning. It combines the characteristics of breadth-first search and heuristic search. It can reduce the search space as much as possible and improve the search efficiency while ensuring the optimal path is found. However, as the map environment becomes more complex, the traditional A* algorithm has problems such as low search efficiency, easy to fall into local optimality and poor path smoothness. Therefore, it is necessary to combine the DQN algorithm to Step S2: extract key points of the global inspection path, and divide the global inspection path into N local inspection paths according to the key points.

[0018] The global inspection path is the entire path of the drone from the starting point to the end point, denoted as L. Because the vision of the drone is also limited like the human eye, it is impossible to see the whole path from the starting point to the end point. Therefore, the initial global inspection path is only a rough path, and it is impossible to know in advance whether there are obstacles in the global planning path. Therefore, according to the drone's field of view, the global inspection path is divided into several local inspection paths, denoted as L 1 , L 2 ......L N Each local inspection path has a local starting point and a local target point. 1 The local target point is L 2 The straight-line distance between the local starting point and the local target point of each local inspection path is less than or equal to the field of view radius of the drone.

[0019] Step S3, the UAV flies along each local inspection path in sequence, monitors the local inspection path where it is located, and determines whether there are random obstacles in the local inspection path where it is located; if not, continue to fly until the local target point of the local inspection path, start monitoring the next local inspection path again, repeat step S3, until it reaches the end of the global inspection path, and end the inspection task; if yes, go to step S4; Here, random obstacles are static obstacles. The global inspection path planned by the A* algorithm cannot plan all static obstacles clearly. Inevitably, during the flight of the drone, there will be some obstacles that are not planned in advance and are on the flight path of the drone. Since the field of view of the drone is limited, it will constantly discover new random obstacles in a static state during the flight.

[0020] The drone monitors the surrounding environment in real time during flight, collects data using sensors, and identifies random obstacles that appear in the local inspection path. The data collected by the sensors includes the location and size of the obstacles.

[0021] Step S4: When random obstacles are detected in the local inspection path, the DQN algorithm is used to replan the local inspection path.

[0022] The A* algorithm mainly performs path planning based on static maps. When the environment changes, such as when obstacles move or when an unknown environment is encountered, if the global path is replanned, this may cause the algorithm to respond more slowly. The flight speed of drones is relatively fast, and re-planning the global path during flight can easily interfere with the flight of drones. If the global path planning is unreasonable, the map update is incomplete, obstacles are not detected, etc., the drone may collide or even explode.

[0023] Therefore, a method needs to be integrated to improve the shortcomings of the A* algorithm. The DQN algorithm has first-class obstacle avoidance capabilities in unknown environments and is based on path planning in dynamic environments. The A* algorithm has the advantages of being fast, efficient, and accurate in global path planning, while the DQN algorithm is good at handling dynamic environments and complex decision-making processes. Combining the two can fully utilize the global planning capabilities of the A* algorithm and the dynamic decision-making capabilities of the DQN algorithm.

[0024] Once a new obstacle is detected, the DQN algorithm is started to replan the local path. DQN can predict the best action strategy based on the current environment state data (including the shape, size and position of the obstacle, and the position of the drone).

[0025] Furthermore, the specific process of replanning the local inspection path using the DQN algorithm is as follows: Take environmental state data as input features; The DQN algorithm identifies environmental status data and outputs action instructions, which include but are not limited to continue moving forward, turn left, turn right, or retreat.

[0026] The drone executes the action instructions output by the DQN algorithm.

[0027] Combination Figure 2 How it works: First, based on the pre-established global map, the A* algorithm is used to generate a map from the starting point N to the global target point G. global The optimal global inspection path between the two. Secondly, the key nodes on the global inspection path are extracted. During the operation of the drone, the current position C of the drone t The nearest path node is selected from the circle with a center and a radius of r as the local target G local , the drone will check the current position C during the movement t With the local target point G local If there are no obstacles, the drone will move along the global optimal path; if there are obstacles, the drone will re-plan the current position C using the DQN algorithm loaded with the pre-trained model. t With the local target point G local until the UAV reaches the global target point.

[0028] Step S5: Whether the end point of the global inspection path is reached, if yes, the inspection task is terminated, otherwise, go to step S3.

[0029] After executing the action determined by the DQN algorithm, the drone updates its position in real time and adjusts the subsequent flight path. The inventor of the present application uses the Gmapping algorithm in the simultaneous localization and mapping (SLAM) technology to map the actual environment and conduct simulation experiments to verify the effectiveness of the hybrid path planning algorithm of the present invention.

[0030] Figures 3 to 6 The following figure shows the process of adjusting the inspection path of the UAV under the hybrid path planning method combining the A* algorithm and the DQN algorithm. Figure 3 is the global inspection path L planned using the A* algorithm. Figure 4 To fly to the local inspection path L k After that, a random obstacle A that originally existed on the inspection path appears. At this time, the DQN algorithm comes into play. By identifying the position and size information of the obstacle, the random obstacle A is avoided and the local inspection path of this section is adjusted to L k ’ , and finally form Figure 6 The present invention only adjusts the local inspection path where random obstacles appear, avoiding adjusting the global inspection path every time a random obstacle is detected, thereby improving the efficiency of path planning.

[0031] Embodiment 2 This embodiment is further improved on the basis of the first embodiment. During the flight of the drone to perform the inspection task, not only static obstacles will appear, but also dynamic obstacles, such as people, animals and railway facilities. Therefore, it is necessary to avoid dynamic obstacles that appear on a certain section of the local inspection path. Therefore, the hybrid path planning method of the drone based on track inspection in this embodiment also includes a method for avoiding dynamic obstacles. The principle is to process the point cloud data generated by the depth camera carried by the drone by combining the YOLOv5 target detection algorithm and the density-based spatial clustering application algorithm (DBSCAN algorithm for short) to achieve effective separation and characterization of obstacles. First, the YOLOv5 target detection algorithm is used to identify areas where obstacles may exist, and then the DBSCAN algorithm further processes the point cloud data in these areas, and clusters the points belonging to the same obstacle together by clustering, thereby achieving separation and characterization of obstacles. After obtaining the clustering results based on the DBSCAN algorithm, the flight speed and direction of the drone are dynamically adjusted according to the type, distance and dynamic characteristics of the obstacle to achieve effective avoidance of dynamic obstacles.

[0032] Specifically, the method for avoiding dynamic obstacles includes: Step 1: The drone reads the image frame data of the depth camera in real time, processes the image frame data using the YOLOv5 target detection algorithm, identifies dynamic obstacles, and quickly and accurately locates and classifies dynamic obstacles.

[0033] The drone stores the names and types of dynamic obstacles that may exist in the railway area.

[0034] The YOLOv5 model extracts features from image frame data through a convolutional neural network (CNN) and predicts the bounding box and category of dynamic obstacles.

[0035] Step 2: Use the DBSCAN algorithm to process the point cloud data generated by the depth camera and group the points belonging to the same obstacle together to form clusters.

[0036] Point cloud data is generated by a depth camera and contains a large amount of point information in three-dimensional space. The DBSCAN algorithm calculates the density relationship between points in the point cloud data and groups points belonging to the same obstacle together to form clusters. For a given point set, closely adjacent points are grouped together under a certain radius and a minimum number of included points. In this way, we can cluster point cloud data belonging to the same obstacle into a cluster. The algorithm does not need to specify the number of clusters in advance, but automatically determines the clustering structure based on the density distribution of the points.

[0037] The DBSCAN algorithm uses the high-density connectivity of classes to quickly discover classes of arbitrary shapes. The basic idea is that for each object in a class, the number of objects contained in its given radius cannot be less than a given minimum number. In order to discover a class, DBSCAN first finds any object B from the database object set A, and then finds the information about R and B in A. min All objects that are density-reachable from B (where R is the radius of B min is the minimum number of objects). If B is a core object, that is, the area of ​​B with a radius of R contains no fewer than B objects. min According to the algorithm, we can find a relationship between the parameters R and B. min If B is a boundary point, then the number of objects contained in the area of ​​B with radius R is less than B min That is, there is no object density-reachable from B. B is temporarily marked as a noise point and then DBSCAN processes the next object in A.

[0038] Step 3: Based on the clustering results of the DBSCAN algorithm and the dynamic obstacles identified by the YOLOv5 target detection algorithm, the points in the point cloud data that belong to the same dynamic obstacle are characterized, and the point cloud data that does not belong to the obstacle is separated.

[0039] The location, shape, and size of dynamic obstacles can be calculated from the point cloud data in the cluster. At the same time, the point cloud data that does not belong to obstacles (such as static background, ground, etc.) is separated to achieve effective separation and characterization of obstacles.

[0040] Step 4: Calculate and analyze the type, distance and movement direction of obstacles, estimate the speed of dynamic obstacles (multi-frame difference + sliding window average), analyze the collision probability, and dynamically adjust the flight speed and direction of the drone to achieve effective avoidance of dynamic obstacles.

[0041] The present invention also provides a storage medium on which a program is installed. When the program is run, the hybrid path planning method for unmanned aerial vehicles based on track inspection of embodiment one or embodiment two can be executed.

[0042] The above is only for explaining the implementation mode of the present invention and is not intended to limit the present invention. For those skilled in the art, any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention without creative work should be included in the protection scope of the present invention.

Claims

1. A hybrid path planning method for UAV based on track inspection, characterized in that: The following steps are involved: Step S1, using the A* algorithm to plan a preliminary global inspection path; Step S2, extracting key points of the global inspection path, and dividing the global inspection path into N local inspection paths according to the key points; Step S3, the UAV flies along each local inspection path in sequence, monitors the local inspection path where it is located, and determines whether there are random obstacles in the local inspection path where it is located; if not, continue to fly until the local target point of the local inspection path, start monitoring the next local inspection path again, repeat step S3, until it reaches the end of the global inspection path, and end the inspection task; if yes, go to step S4; Step S4: when a random obstacle is detected in a local inspection path, the DQN algorithm is used to replan the local inspection path; Step S5: Whether the end point of the global inspection path is reached, if yes, the inspection task is terminated, otherwise, go to step S3.

2. The hybrid path planning method for UAV based on track inspection according to claim 1 is characterized in that: It also includes methods for avoiding dynamic obstacles: The drone reads the image frame data of the depth camera in real time, processes the image frame data using the YOLOv5 target detection algorithm, identifies dynamic obstacles, and quickly and accurately locates and classifies dynamic obstacles; Use the DBSCAN algorithm to process the point cloud data generated by the depth camera and group the points belonging to the same dynamic obstacle together to form a cluster; Based on the clustering results of the DBSCAN algorithm and the dynamic obstacles identified by the YOLOv5 target detection algorithm, the points in the point cloud data that belong to the same dynamic obstacle are characterized, and the point cloud data that does not belong to the dynamic obstacle is separated; Calculate and analyze the type, distance and movement direction of dynamic obstacles, estimate the speed of dynamic obstacles, analyze the collision probability, and dynamically adjust the flight speed and direction of the drone to achieve effective avoidance of dynamic obstacles.

3. The hybrid path planning method for UAV based on track inspection according to claim 1 or 2 is characterized in that: The specific process of replanning the local inspection path using the DQN algorithm is as follows: Take environmental state data as input features; The DQN algorithm outputs action instructions by identifying environmental state data; The drone executes the action instructions output by the DQN algorithm.

4. The hybrid path planning method for UAV based on track inspection according to claim 3 is characterized in that: The environmental state data includes the shape, size and position of random obstacles, and the position of the drone.

5. The hybrid path planning method for UAV based on track inspection according to claim 3 is characterized in that: The action instructions include but are not limited to continue moving forward, turning left, turning right or retreating.

6. The hybrid path planning method for UAV based on track inspection according to claim 2 is characterized in that: Using the DBSCAN algorithm to group points belonging to the same dynamic obstacle together to form clusters includes: the DBSCAN algorithm calculates the density relationship between points in the point cloud data, and for a given point set, with a certain radius and a minimum number of included points as conditions, classifies closely adjacent points into one category, and groups points belonging to the same dynamic obstacle together to form clusters.

7. A storage medium having a program installed thereon, characterized in that: Running this program can execute the hybrid path planning method for UAV based on track inspection as described in any one of claims 1 to 6.

Images