Research On Binocular Vision SLAM Of UAV In Unknown Environment

With the widely used of Unmanned Aerial Vehicle(UAV)in the area of military and civil,achieving the autonomous navigation of UAV is the topic.In the absence of GPS and without the previous information about the environment,namely,under the unknown environment,the key of indoor and outdoor autonomous navigation is Simultaneous Localization and Mapping(SLAM).Cameras are into the field of SLAM,because of its cheap and capturing large information from the environment.The theme of thesis is the SLAM of UAV based on binocular vision.This thesis analyzes the basic theory of SLAM based on binocular vision,and conducts relative experiments.First of all,SIFT and ORB which are the two feature extraction algorithms of image are analyzed theoretically.The hardware platform of SLAM for UAV based on binocular vision is built.Both indoor and outdoor,the experiments based on the SIFT and ORB are carried out.As a result,ORB takes less time than SIFT,otherwise,the robustness of SIFT is stronger than ORB.Then,the front end of binocular vision SLAM based on feature is accomplished.In the situation of indoor and outdoor,the experiment of Visual Odometry(VO)is carried out.The features still are SIFT and ORB.As a result,the initial position and pose are got.So the local 3D point cloud map is obtained.The result of experiment indicates that the VO based on SIFT has an advantage of the accuracy in the estimation of position outside,comparing with the VO based on ORB.Both are well-matched indoors.However,the accuracy of VO based on ORB is higher than VO based on SIFT.In addition,the back end of binocular vision SLAM based on feature is completed.The experiment of SLAM based on binocular vision is carried out.Under the conditions that keyframe and Loop Closure are considered,the Nonlinear Optimization with pose graph is used for the optimization of the initial position and pose.The experimental result shows that the position of binocular vision SLAM and map are more accuracy than VO.Finally,the SLAM of UAVs based on binocular vision is studied.Two schemes are studied and applied to two UAVS based on binocular vision SLAM.The simulation of two UAVS based on binocular vision SLAM is carried out.The simulation experiments from the second schemes show that the SLAM of two UAVs is not only better than the SLAM of single UAV in the accuracy of position and map,but takes less time.