Stereo Vision And Inertial Navigation Fusion Oriented Position And Attitude Estimation Method For UAV

Nowadays,QUAVs are more and more widely used and intelligent in various fields.The autonomous navigation in unknown environment is the key to realize intelligence,and the core of autonomous navigation is positioning technology.Therefore,the positioning problem of UAV in unknown environment is studied.In addition,compared with terrestrial robots,UAV positioning requires higher accuracy and robustness.The improved visual odometer method is studied to improve positioning accuracy.Aiming at the tracking failure of positioning system caused by too fast motion and rotating motion,the inertia and vision tightly coupled navigation system based on non-linear optimization is studied to improve the robustness of the system.The main contents of this paper are as follows:Firstly,the visual odometer is mainly studied.In order to improve the accuracy and speed of pose estimation,a series of improvements to the visual odometer are proposed.Firstly,in feature matching period,a feature point preprocessing method based on DL index is proposed to select feature points with high reliability;secondly,a feature matching algorithm based on IMU prior knowledge is proposed to improve matching speed;After that,according to the difference of pose accuracy caused by different camera motion speed,a feature point extraction method based on key frame construction speed is proposed.Finally,according to the drift problem of spatial points in high-speed motion or rotational motion,a pose estimation method based on bidirectional reprojection is proposed.Secondly,the visual INS algorithm is mainly focused.In order to enhance the robustness of the algorithm on UAV,the tightly coupled optimization method of vision and IMU based on sliding window is adopted.In order to improve the real-time performance of the algorithm on embedded devices,an improved Bundle Adjustment solution method is proposed.Thirdly,the experimental platform of DJI M100 UAV is built by using Nvidia TX2 developer kit and DJI Guidance visual navigation system,and the experimental verification of the algorithm is carried out.The experimental results are compared with ORB-SLAM2 to verify the superiority of the algorithm in accuracy and robustness.Compared with ORB-SLAM2,the simulation and UAV experiment results show that the proposed algorithm can suppress cumulative errors better and improve the accuracy of the system.In rotational motion,the proposed algorithm performs better than ORB-SLAM2,and the robustness of the system is enhanced.