UAV Path Planning Based On Visual Slam Mapping

With the upsurge of artificial intelligence research,Unmanned Aerial Vehicle(UAV)autonomous flight has received extensive attention and research.To achieve UAV autonomous flight,first of all,two basic key technologies are required,one is three-dimensional mapping,and the other is path planning.Simultaneous Localization and Mapping(SLAM)technology is a very commonly used core technology for constructing environmental maps,which is of great significance for UAV path planning and autonomous flight.In this project,a depth camera is used as a sensor to measure the environment,and a SLAM system capable of constructing a high-precision map is designed.Path planning is the basis for UAV navigation.RRT algorithm is widely used with extremely high flexibility.However,the convergence speed of the RRT algorithm decreases in the three-dimensional space,and the planned path cannot guarantee the optimal.In order to make up for these shortcomings of the RRT algorithm,this topic has made improvements and applied the improved RRT algorithm to the three-dimensional UAV Path planning to achieve autonomous flight of drones.The main research contents of the subject are as follows.1.Using ORB features,an RGB-D SLAM mapping system based on multithreading is proposed.For the extraction of ORB features,it is proposed to improve the image by region extraction,so that the extracted feature points can be evenly distributed on the image.This improvement can effectively reduce the number of feature mismatches and improve the robustness of matching.Furthermore,it proposes the improvement of the numbering of the image feature points,which accelerates the speed of feature matching between the image and the local map,and ensures the real-time performance of the visual odometer.2.The global back-end optimization based on non-linear graph optimization is added to the design visual odometer,so that the camera pose estimated by the front-end at each moment is optimized to the greatest extent.Subsequently,a new thread was added to construct a dense point cloud map,so that the SLAM system designed by the subject can build a high-precision dense point cloud map in real time.3.An improved RRT algorithm was proposed,which improves the traditional RRT algorithm from a two-dimensional algorithm to a three-dimensional planning algorithm.In view of the shortcomings of the RRT algorithm,the improvement of the heuristic factor combining the direction of force as the growth direction of the random tree is proposed.The IRRT algorithm can make the growth direction of the random tree always toward the vicinity of the target point,thereby reducing unnecessary exploration of space by the algorithm,and greatly reducing the planning time of the algorithm.Because the growth direction of the random tree is determined,the number of nodes included in the path planned by the algorithm is effectively reduced,so that the planned path tends to be optimal.Finally,a method for path smoothing is proposed.After smoothing the path using this method,the quality of the resulting path will be better.The experimental results prove the superiority of the algorithm.