Research And Implementation Of UAV Visual Inertial Positioning System

UAVs usually employ Global Position System(GPS)for accurate positioning in the industrial scenario,such as post-disaster search and rescue,bridge maintenance,power inspection and urban patrol.If there is disturbance on compass,GPS measurement cannot be utilized effectively,which will lead to no capability of precision localization.To this end,we research employing Visual Inertial Odometry(VIO)algorithm to assist precise positioning of UAVs,which only requires a small and lightweight sensor package and performs well in full texture scenes.Optimization-based VIO usually has high overhead and poor robustness of initialization Even though,the VIO based on Multi-State Constraint Kalman Filter(MSCKF)has low overhead,it contains the dreadful performance of initialization and State estimation,and cannot be employed on high-altitude environment.In this paper,the advantages of MSCKF are combined,and a series of improvement strategies on original MSCKF framework are implemented.The resulting VIO scheme with high robustness,low overhead and application in high-altitude environment is designedIn this paper,we mainly discuss the theories of VIO algorithm in detail from aspects of front-end visual processing,back-end feature point estimation and state update.And we correct the wrong sub-block of covariance augmentation Jacobian matrix in original MSCKF In practice,we formulate zero speed detection and Zero Velocity Update(ZUPT)of Inertial Navigation System(INS)into the back-end of VIO to overcome static cases as well as to improve initialization of MSCKF to support both motional and static initialization.In addition,we formulate a high-end incremental nonlinear optimization method to achieve MSCKF performs well at high altitudes.we evaluate our method in real-world data-sets and compare our method with other state-of-the-art VIO in terms of overhead,accuracy,and robustness.It demonstrates that our method performs better.In the end,the experiment of UAV flying with VIO in GPS-denied scenes is carried out,and it works well.It demonstrates that our method has practical application value.