Research On Pose Estimation Based On Artificial Features For MAV

With the rapid development in recent years,Unmanned Aerial Vehicles(UAVs)constantly play increasingly crucial roles on users' s life.The UAV's autonomousness is,however,a huge challenge.UAV's pose estimation is the core towards its autonomousness.This paper aims at research on pose estimation for UAV.In order to make the system more reliable,special tag is used as the artificial features.Two methods have been proposed based on Extended Kalman Filter(EKF)and Nonlinear Least Square(NLS)respectively.The algrithms are validated,evaluated and compared by simulation sensor data and real world sensor data.Firstly,pinhole camera model is selected as the camera observation model.The model is a bridge between 3D point in the real world and features in image plane.It is essential for the following estimation methods based on EKF and NLS.Then,the tag detection algorithm is introduced briefly and the transformation of the tag frame relative to the camera frame can be obtained by the solution of the perpective N point problem.Next,the method based on EKF is studied.Firstly,the noise model and the continuous kinematic model of the MEMS IMU are investigated.The prediction of IMU's position,velocity and attitude can be obtained via integration of continuous kinematic model.The indirect Kalman filter is employed and the error dynamics is deduced to update error covariance.The four corners' pixel coordinates of detected tag are selected as an observation.Finally,the simulation and the real world sensor data are used to evaluate the algorithm's accuracy.Finally,pose estimation based on NLS is studied.Firstly,the general idea of NLS for estimation is clearified: defining the state,exploring the observation model to obtain the residual and minmizing the objective function for optimized states.Then,two residual functions are defined according to the sensor types: the residual of the IMU preintegration measurements and the residual of tag image pixel coordinates.Then the relationship between the residues and the states are investigated.Finally,the simulation and real world data are employed to verify the algorthm and compared to the EKF-based method.