Research On UAV Obstacle Detection And Path Planning Algorithm

With the wide application of low-cost multi-rotor UAVs in the military and civilian fields,the requirements for environmental perception and path planning capabilities of multi-rotor UAVs have also increased.In this paper,we will study obstacle detection and path planning from UAV,and realize the detection of obstacles in the images acquired by the camera through the improved YOLOv4 deep learning algorithm.Then,according to the detection results and the principle of binocular vision,the relative distance from the obstacle is obtained.Finally,according to the situation of environmental perception,the UAV is controlled to carry out obstacle avoidance operations.The contents of this article are as follows:Firstly,a lightweight detection algorithm is proposed for the characteristics of limited computing resources of UAV.In the original YOLOv4,the Mobile Net network and the effective channel attention mechanism are integrated to reduce the memory occupation of the network model.The residual structure fusion module is introduced to enhance the feature extraction capability of the network.Experiments show that the average detection accuracy of the algorithm reaches 80.64%,the memory proportion of the improved detection algorithm is reduced by 80%,and the detection speed reaches 49.21 frames per s,which can meet the real-time requirements of UAVs.Then,according to the detection results of improved YOLOv4,the redundant information is removed,and then F-ORB is used to complete the feature matching of the obstacle area,which reduces the complexity of the calculation and improves the matching accuracy.Secondly,the D-PSO algorithm is proposed when the global flight environment information is known.Firstly,the Dijkstra algorithm is used to complete the shortest path search in a small range,and then the global search path rate is improved by combining the improved particle swarm algorithm.Aiming at the problem that local optimization is easy to fall into in particle swarm arithmetic,the update speed method is optimized and the inertia weight is adaptively updated.The cubic B-spline interpolation algorithm is used to smooth the route.Finally,3D simulation experiments show that the proposed algorithm improves both convergence and route smoothness.Next,the D-PSO algorithm is proposed when the global flight environment information is known.Firstly,the Dijkstra algorithm is used to complete the shortest path search in a small range,and then the global search path rate is improved by combining the improved particle swarm algorithm.This method constructs a multi-factor fitness function,optimizes the update speed method,introduces the adaptive update inertia weight for the problems that are easy to fall into the local optimal value in particle swarm algorithm.In order to prevent sudden changes in speed,the cubic B-spline interpolation algorithm is selected to smooth the route.Finally,3D simulation experiments show that the proposed algorithm improves both convergence and route smoothness.Finally,a local obstacle avoidance strategy based on the improved artificial potential field method is proposed.In order to avoid falling into the phenomenon of local optimal and unreachable target points,this paper introduces virtual target points and adds correction factors to the repulsive field function.Finally,the feasibility of improving the artificial potential field method is verified in MATLAB.At the same time,a simulation environment is built in the Gazebo simulation platform to verify the practicality of the vision-based obstacle detection and obstacle avoidance algorithm.