Research On Obstacle Avoidance Algorithm Of Unmanned Vehicle Based On Multi-sensor Fusion

With the increasing application of unmanned vehicles,more and more types of sensors are integrated in unmanned vehicles.The fusion and unification of the detection results of different sensors can improve the redundancy and fault tolerance of the unmanned vehicle system,thus ensuring the accuracy of obstacle avoidance.In this article,the multi-sensor fusion algorithm and obstacle avoidance algorithm based on path planning are studied for the obstacle avoidance problem of multi-sensor unmanned vehicles,the specific research contents are as follows:Firstly,sensor measurement principles and object detection algorithms are studied.The measurement principles of different sensors are compared,and the detection results of lidar and camera are fused.The ray ground filter method is used to remove the ground points.The method of constructing Kdtree is used to speed up search efficiency of the Euclidean clustering for point clouds,and the dynamic threshold method is used to improve the Euclidean clustering effect;Distortion correction is first performed on the image to ensure image authenticity,and then the real-time YOLOv5 algorithm is used to achieve camera detection of the external environment.The feasibility of the single sensor target detection algorithm is verified by the simulation test.Secondly,a multi-sensor fusion algorithm based on target tracking stability is designed.Firstly,the coordinate transformation relationship between lidar and camera in the open source dataset KITTI is deduced,and is applied to the program to realize the mapping of point cloud coordinates to images,the detection results of different sensors are unified;Secondly,the SORT algorithm combining Kalman filter and Hungarian matching is studied to track the target detection results of the single sensor.The weight of the target detection results of different sensors is allocated through the statistics of the Kalman filter covariance data changes in the process of target tracking.The stability of the unmanned vehicle perception system is improved by using the weight distribution mechanism and the sensor exit mechanism.Simulation results show that the multi-sensor fusion algorithm with dynamic weight allocation unifies the detection results of different sensors,so that the unmanned vehicle can obtain more comprehensive and accurate obstacle information.Finally,a two-layer unmanned vehicle obstacle avoidance algorithm based on improved DWA is designed.Firstly,the A * algorithm is used to achieve global path planning for unmanned vehicles on known map;Secondly,a local path planning(DWA)algorithm based on multi-sensor obstacle detection results is studied.In response to the problems of unsmooth paths,circumnavigating the area of dense obstacles,and local minima in the DWA algorithm,a method of selecting pre trajectories based on manifold space is adopted to improve the smoothness of the path planning process for unmanned vehicles;Obstacle evaluation subfunction is added to solve the problem of DWA algorithm bypassing areas with dense obstacles;Azimuth evaluation subfunction is improved to enable unmanned vehicles to run smoothly even when approaching the target point;A dual-layer obstacle avoidance algorithm is designed by integrating the A~* algorithm to solve the local minimum problem.The simulation results show that the two-layer unmanned vehicle obstacle avoidance algorithm based on A~* and improved DWA has shorter path length and less time compared to the previous algorithm,which improves the working efficiency of unmanned vehicles and make vehicle reach the target point smoothly.