Micro-UAV Onboard Geo-pose Estimation On A Ground Moving Target

Recently,there is a trend of intelligence and swarm for UAV technology development,which poses challenges for UAV tracking and localizing ground moving targets.Intelligence requires UAV perform operations in real time on the onboard processors with limited computing resources;while UAV swarm requires a unified framework for each single UAV to facilitate information sharing for each other.Under this circumstance,this dissertation focuses on the problem of estimating the ground target’s geo-pose with a UAV.The main key technologies include online geo-pose estimation,estimating metric position of moving target,state filtering,trajectory smoothing.The main work and contribution are summarized as follows:(1)The problem of UAV tracking ground moving targets is theoretically modeled,and the target localization in the inertial frame is realized.The high-speed movement of the UAV and the uncertainty movement of the ground target cause the object frame and the camera frame to be constantly changing.A scheme of visual measurement and combination with self-positioning of onboard sensors is proposed to estimate the target position in geodetic frame.According to the overall scheme,the error sources are analyzed and the corresponding optimization solution is proposed to improve localization accuracy.(2)An extrinsic calibration and optimization method for onboard multi-sensors system is proposed.We proposed a multi-sensors extrinsic calibration method based on the nonlinear optimization.Different from the calibration board aided methods,this method acquires image and drone data through flight experiments and establishes the correspondence between the image and physical space.Then use the nonlinear optimization method to minimize the error between the estimation value and real value to find the optimal parameter.Real flight experiments demonstrate the effectiveness of the algorithm.(3)An iterative filtering and local weighted regression-based online smoothing method for ground target motion trajectory are proposed.Considering the requirements of online operations with a lightweight onboard processor,a recursive filtering method based on extended Kalman Filter(EKF)is used to estimate the pose of the target.And then the dynamic error of EKF and the nonconvergence of EKF are analyzed.Based on the results of discrete sampling and filtering,a local weighted regression method is proposed to smooth target trajectory.This method only uses the past data to fit the trajectory of the target for a period of time.