Unmanned aerial vehicle obstacle avoidance system

Abstract

An unmanned aerial vehicle obstacle avoidance system disclosed by the present invention comprises a hardware unit and a software unit, the hardware unit comprises an unmanned aerial vehicle body, an upper computer, a sensor and a ground station, and the software unit comprises an obstacle avoidance algorithm, an A * algorithm, an RRT algorithm, a radar data sequence and point cloud data. The unmanned aerial vehicle body comprises a sensor, an upper computer and a flight control module, and the unmanned aerial vehicle body is connected with the ground station. According to the method, the unmanned aerial vehicle body and the ground station are split, a new algorithm needing to be used can be deployed on the ground station or an existing algorithm can be modified to cope with path planning in a complex environment, and path planning is performed according to information obtained by the unmanned aerial vehicle body through a sensor. Data acquired by the laser radar and the depth camera are directly fused and converted into laser radar data for obstacle avoidance of the unmanned aerial vehicle, so that the laser radar and the depth camera can be better matched to realize own functions instead of constructing an obstacle avoidance effect directly through the unmanned aerial vehicle.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle paths, and in particular to an unmanned aerial vehicle obstacle avoidance system. Background Art

[0002] Under the current technical environment, the autonomous obstacle avoidance capability of drones mainly relies on onboard sensors and specific obstacle avoidance algorithms. After the drone is started, it collects data about the surrounding environment through sensors and transmits this data to the onboard computer system. This system is responsible for analyzing and processing this information and planning a safe flight route in combination with the obstacle avoidance algorithm. Subsequently, the computer converts the planned path into specific flight instructions, guiding the drone to fly autonomously along the established path until it reaches the destination. The accuracy of autonomous obstacle avoidance of drones is affected by many factors, including the processing performance of the onboard computer, the accuracy of the sensor, the path planning accuracy of the obstacle avoidance algorithm, the stability of the communication, and the external environmental conditions. Among them, the performance of the onboard computer, the quality of the sensor, and the accuracy of the obstacle avoidance algorithm are key factors that can be optimized and adjusted manually. In order to improve the performance of drones in autonomous obstacle avoidance, efficient coordination between these three elements is required.

[0003] The current autonomous obstacle avoidance technology for drones generally relies on a single obstacle avoidance algorithm to achieve automatic obstacle avoidance. However, due to the performance limitations of sensors and the limited processing power of onboard computers, the actual effect of this obstacle avoidance method is often not ideal. These problems are not only restricted by hardware conditions, but are also easily affected by changes in the external environment, thus posing a severe challenge to the safe and autonomous flight of drones. Obviously, there is significant room for improvement in the current autonomous obstacle avoidance technology for drones, and it is urgent to improve its obstacle avoidance effect and flight safety in complex environments through technological innovation. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a drone obstacle avoidance system, the purpose of which is to solve the problems of large resource consumption and high hardware requirements in the process of autonomous obstacle avoidance of drones, and to improve the accuracy and stability of autonomous obstacle avoidance of drones.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect of the present invention, a drone obstacle avoidance system is provided, comprising a hardware unit and a software unit, wherein the hardware unit comprises a drone body, a host computer, a sensor and a ground station, and the software unit comprises an obstacle avoidance algorithm, an A* algorithm, an RRT algorithm, a radar data sequence and point cloud data; the drone body comprises a sensor, a host computer and a flight control module, wherein:

[0007] The drone body is used to connect to the ground station, which contains a software unit. The software unit based on the ground station provides the drone body with path planning and obstacle avoidance algorithms. In the drone body, the sensor is used to obtain environmental information to generate rasterized map information of environmental information and related obstacle features. The host computer is used to read the sensor data and transmit the rasterized map information to the ground station through the data network. The ground station is used to parse the rasterized map information transmitted by the host computer and transmit it back to the drone body after processing according to the software unit.

[0008] In the ground station, the A* algorithm performs global path planning based on the rasterized map information to generate the basic path. The RRT algorithm is used to optimize the path in areas with many obstacles based on the A* algorithm, with the path point before the drone enters the area as the local starting point and the path point when the drone leaves the area as the local end point, thus forming a secondary calculation of the path planning.

[0009] As a preferred technical solution of the present invention, the A* algorithm also includes the following operation process:

[0010] (1) Reduce the number of path points in the path and smooth the path. Specifically, start from the starting point, read three adjacent points continuously, and determine whether there is an obstacle in the straight line connected by the first point and the third point and the area with a safe distance radius around the midpoint of the connected straight line. If there is no obstacle, delete the second point and replace the second point with the third point. Execute this logic in sequence until you reach the end point.

[0011] (2) Improve the node search logic, take the drone body as the central node, and expand the current node to eight directions: left, upper left, lower left, upper, lower, right, upper right, and lower right. Specifically: (a) During the search, if there is an obstacle in one of the directions, it is recorded as an obstacle point. The three points in the adjacent directions are moved out of the search range, and the node with the lowest cost is reselected. The other points farthest from the obstacle point are selected according to the diagonal principle; (b) When there are no obstacle points on the diagonal lines, if there is an obstacle on one side of the straight line of the central node, the point is abandoned and the point with the lowest cost along the straight line is selected as the next node.

[0012] As a preferred technical solution of the present invention, the sensor further includes a laser radar and a depth camera, which are used to generate a radar data sequence and point cloud data respectively. The fusion process of the radar data sequence and the point cloud data is as follows:

[0013] S1. Calibrate the internal parameters of the depth camera and calculate the external parameter data of the depth camera and the lidar, downsample the point cloud data of the depth camera, remove the point cloud data that is too high and on the ground, and pseudo-process the pre-processed point cloud data of the depth camera into lidar data;

[0014] S2. Convert the LiDAR data into point cloud format, filter out isolated noise points without destroying the point cloud structure, and remove duplicate and erroneous point clouds;

[0015] S3. Use the sensor external parameter matrix to transfer the simulated laser point cloud to the lidar coordinate system through coordinate transformation, fuse the two point cloud data in the above coordinate system, and convert the fused point cloud data into lidar data through a program, and finally output it to the ground station to form the basis for path planning.

[0016] In a second aspect of the present invention, a computer device is provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method steps or system functions of the first aspect when executing the computer program.

[0017] In a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method steps or system functions of the first aspect are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1: The present invention separates the drone body and the ground station, and can deploy new algorithms or modify existing algorithms at the ground station to cope with path planning in complex environments. Based on the information obtained by the drone body through sensors, the data obtained by the lidar and depth camera are directly fused and converted into lidar data for drone obstacle avoidance, so that the two can better cooperate to realize their own functions, rather than directly constructing obstacle avoidance effects through the drone itself.

[0020] 2: The present invention further optimizes the problem of low path planning accuracy. It adopts the A* algorithm for global planning and the RRT algorithm for local complex environments, and optimizes the solution idea of ​​the A* algorithm logic, so that path planning can construct the effect of quadratic planning. In addition, the A* algorithm constrains the calculation idea, which can better save UAV resource consumption and increase the hovering time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0022] Figure 1 It is a schematic diagram of the overall flow of the system of the present invention;

[0023] Figure 2 It is a flow chart of the fusion of sensor radar data sequence and point cloud data of the present invention. DETAILED DESCRIPTION

[0024] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0025] In this embodiment, if Figure 1-2 As shown, the present invention provides a drone obstacle avoidance system, including a hardware unit and a software unit, wherein the hardware unit includes a drone body, a host computer, a sensor and a ground station, and the software unit includes an obstacle avoidance algorithm, an A* algorithm, an RRT algorithm, a radar data sequence and point cloud data; the drone body includes a sensor, a host computer and a flight control module, and the drone body is connected to the ground station, wherein:

[0026] The drone body is used to connect to the ground station, which contains a software unit. The software unit based on the ground station provides the drone body with path planning and obstacle avoidance algorithms. In the drone body, the sensor is used to obtain environmental information to generate rasterized map information of environmental information and related obstacle features. The host computer is used to read the sensor data and transmit the rasterized map information to the ground station through the data network. The ground station is used to parse the rasterized map information transmitted by the host computer and transmit it back to the drone body after processing according to the software unit.

[0027] In the ground station, the A* algorithm performs global path planning based on the rasterized map information to generate the basic path. The RRT algorithm is used to optimize the path in areas with many obstacles based on the A* algorithm, with the path point before the drone enters the area as the local starting point and the path point when the drone leaves the area as the local end point, thus forming a secondary calculation of the path planning.

[0028] Specifically, the working principle of the data transmission part of the drone, ground station, and host computer is as follows: after the drone body takes off, the sensor starts working, obtains environmental information and generates data, the host computer reads the sensor data and packages it, and sends the data from the host computer to the ground station through the local area network. The ground station parses the acquired data packet, and the ground station generates the map used after processing the acquired data, and executes the obstacle avoidance algorithm, generates the flight path, and generates flight instructions according to the position of the drone body, and sends the flight instructions to the host computer through the local area network. The host computer is connected to the flight control module through a wire, and the flight control module controls the actual flight of the drone according to the instructions; the drone body itself can also be equipped with a separate obstacle avoidance module, such as an image obstacle avoidance algorithm within a range of 2m, which is only used for real-time obstacle avoidance at close range. For long-distance path planning and obstacle avoidance, it is still completed through the ground station, thereby reducing the power consumption of the drone body and increasing the computing speed of the drone body.

[0029] The working principle of the obstacle avoidance algorithm is as follows: on the map established based on environmental information and the location of the drone body, set the target point, use the A* algorithm for global path planning, and after generating a basic path, use the RRT algorithm for the path in the area with more obstacles, with the path point before the drone body enters the part as the local starting point, and the path point leaving the part as the local end point, run the RRT algorithm, optimize the path based on the generated path, ensure that the drone maintains a certain safe distance from the obstacle, and avoid collision. In this way, the optimized path is obtained, the generated path is smoothed, large-angle turns are reduced, and nodes are optimized, so as to obtain a safe and efficient flight path.

[0030] The specific fusion process is as follows: through the mapping of the environment by sensors, a real environment map is established based on environmental information and relevant obstacle features, and the map is rasterized. The global path planning A* algorithm is used as the first layer, and the local path planning RRT method is used as the second layer of operation. First, the first layer of operation is entered to perform the first operation to generate the global original path, and the turning points in the path are stored in the sub-target point collection. Enter the second layer of operation, and use the RRT algorithm to perform the second operation on all stored sub-target points. The turning points in the global original path are processed to achieve the goal of avoiding obstacles.

[0031] The improvements of the A* algorithm are mainly in two aspects:

[0032] (1) Reduce the number of path points in the path and smooth the path. Specifically, starting from the starting point, read three adjacent points continuously, determine whether there are obstacles in the straight line connected by the first point and the third point and in the area with a safe distance radius around the midpoint of the connected straight line. If there are no obstacles, delete the second point and replace the second point with the third point. Execute this logic in sequence until the end point is reached.

[0033] (2) Improve the node search logic, specifically: (a) When searching in the eight directions (left, upper left, lower left, upper, lower, right, upper right, lower right) of the current node, if there are obstacles on the right and lower sides at the same time, the three points on the right, lower and lower right should be removed from the search range and the node with the lowest cost should be reselected. Similarly, the three situations of right, upper, upper right; left, lower, lower left; left, upper, upper left should be selected according to the above logic, that is, when there are obstacles on both sides of the four diagonals of the node, the diagonal direction should be abandoned and the node with the lowest cost should be reselected as the next node. (b) When searching in the eight directions of the current node, if there is a minimum cost point on the upper left, lower left, upper right, or lower right, if there is an obstacle on one side of the connected straight line, the point should be abandoned and the minimum cost point along the straight line should be selected as the next node.

[0034] Through the above two improvements, the number of nodes selected by the algorithm is reduced by 21%, and the number of inflection points generated is reduced by 12%. It can complete path planning in complex environments over long distances. If there are close obstacles, the drone still relies on its own close-range obstacle avoidance algorithm to complete temporary obstacle avoidance operations.

[0035] The principle of the sensor data fusion process is as follows: LiDAR and depth camera are mainly used to generate radar data sequence and point cloud data respectively. The internal parameters of the depth camera need to be calibrated and the external parameter data of the depth camera and the LiDAR need to be calculated. The point cloud data of the depth camera is downsampled to remove the point cloud data that is too high and on the ground, and the pre-processed point cloud data of the depth camera is pseudo-processed into LiDAR data. The LiDAR data is then converted into a point cloud format, and isolated noise points are filtered out without destroying the point cloud structure, and repeated and erroneous point clouds are removed. The simulated laser point cloud is transferred to the LiDAR coordinate system through coordinate transformation using the sensor external parameter matrix, and the two point cloud data are fused in the above coordinate system. The fused point cloud data is converted into LiDAR data through a program to provide support for the UAV path planning algorithm, so that it can be used for UAV path planning.

[0036] In another embodiment, a computer device is provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method steps or system functions of other embodiments when executing the computer program.

[0037] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method steps or system functions of other embodiments are implemented.

[0038] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An unmanned aerial vehicle obstacle avoidance system, comprising a hardware unit and a software unit, characterized in that: The hardware unit includes the drone body, host computer, sensor and ground station, and the software unit includes obstacle avoidance algorithm, A* algorithm, RRT algorithm, radar data sequence and point cloud data; the drone body includes sensors, host computer and flight control module, among which: The drone body is used to connect to the ground station, which contains a software unit. The software unit based on the ground station provides the drone body with path planning and obstacle avoidance algorithms. In the drone body, the sensor is used to obtain environmental information to generate rasterized map information of environmental information and related obstacle features. The host computer is used to read the sensor data and transmit the rasterized map information to the ground station through the data network. The ground station is used to parse the rasterized map information transmitted by the host computer and transmit it back to the drone body after processing according to the software unit. In the ground station, the A* algorithm performs global path planning based on the rasterized map information to generate the basic path. The RRT algorithm is used to optimize the path in areas with many obstacles based on the A* algorithm, with the path point before the drone enters the area as the local starting point and the path point when the drone leaves the area as the local end point, thus forming a secondary calculation of the path planning.

2. The obstacle avoidance system for unmanned aerial vehicles according to claim 1, characterized in that: The A* algorithm also includes the following operation flow: (1) Reduce the number of path points in the path and smooth the path. Specifically, start from the starting point, read three adjacent points continuously, and determine whether there is an obstacle in the straight line connected by the first point and the third point and the area with a safe distance radius around the midpoint of the connected straight line. If there is no obstacle, delete the second point and replace the second point with the third point. Execute this logic in sequence until you reach the end point. (2) Improve the node search logic, take the drone body as the central node, and expand the current node to eight directions: left, upper left, lower left, upper, lower, right, upper right, and lower right. Specifically: (a) During the search, if there is an obstacle in one of the directions, it is recorded as an obstacle point. The three points in the adjacent directions are moved out of the search range, and the node with the lowest cost is reselected. The other points farthest from the obstacle point are selected according to the diagonal principle; (b) When there are no obstacle points on the diagonal lines, if there is an obstacle on one side of the straight line of the central node, the point is abandoned and the point with the lowest cost along the straight line is selected as the next node.

3. The obstacle avoidance system for unmanned aerial vehicles according to claim 2, characterized in that: The sensor also includes a laser radar and a depth camera, which are used to generate a radar data sequence and point cloud data respectively. The fusion process of the radar data sequence and the point cloud data is as follows: S1. Calibrate the internal parameters of the depth camera and calculate the external parameter data of the depth camera and the lidar, downsample the point cloud data of the depth camera, remove the point cloud data that is too high and on the ground, and pseudo-process the pre-processed point cloud data of the depth camera into lidar data; S2. Convert the LiDAR data into point cloud format, filter out isolated noise points without destroying the point cloud structure, and remove duplicate and erroneous point clouds; S3. Use the sensor external parameter matrix to transfer the simulated laser point cloud to the lidar coordinate system through coordinate transformation, fuse the two point cloud data in the above coordinate system, and convert the fused point cloud data into lidar data through a program, and finally output it to the ground station to form the basis for path planning.

4. A computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method steps of claim 3 or the system functions of claims 1-2 when executing the computer program.

5. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method steps described in claim 3 or the system functions described in claims 1-2.

Images