Research And Implementation Of Vision Based Autonomous Obstacle Avoidance System For A Quadrotor

In recent years,for the frequent occurrence of quadrotor collision problems,a lot of scientific research institutions and companies at home and abroad have successively conducted research on the obstacle avoidance of quadrotors.With the rapid development of Computer Vision,the combination of computer vision and quadrotor has more and more research meaning and value.Based on it,in this paper a quadrotor platform is set up,which completed the positioning,environmental reconstruction and the obstacle avoidance motion planning of the quadrotor based on the ZED camera.This paper mainly focuses on the research and engineering of the autonomous obstacle avoidance based on Computer vision.It mainly includes research and engineering of the quadrotor platform,design and implementation of simultaneous localization and mapping based on Computer Vision,motion planning with obstacle avoidance.This paper completed the construction of the quadrotor hardware platform.First,the basic principles of the quadrotor are studied and introduced,including the basic composition,coordinate system,control system,etc.And then the patrol flight is realized based on the Pixhawk module.Based on the construction of the quadrotor hardware platform,this paper completed the positioning of the quadrotor and the reconstruction of three-dimensional dense maps through visual SLAM.The main work of this part is to homogenize the extraction of ORB feature points,avoiding the problem of concentrated feature points extraction in the image;the matching results were filtered twice to get accurate matching results,and the camera pose was obtained through BA estimation.Optimized the pose of the camera and spliced the point cloud to obtain the point cloud map,and finally completed the reconstruction of the dense map.On this basis,this paper completed the modeling and collision detection of the quadrotor and the flight environment.Then realized motion planning which meeting flight performance constraints and obstacle avoidance through an improved Rapidly Exploring Random Tree Star algorithm.After completed the plan,the status of next time will be sent to the lower machine to lead the quadrotor to complete the obstacle avoidance flight and arrive at the destination.Finally,a series of experiments were designed,including the performance test of the quadrotor platform and the performance test of obstacle avoidance system based on Computer Vision.The function and stability of the autonomous obstacle avoidance system based on the Computer Vision were verified.At last,the summary and future work prospect of this paper were also given.