Vision based navigation system of UAV and improvements of the corresponding filtering algorithm

A vision/inertial integrated navigation system was built. The corresponding filtering model was established by treating the motion models of aerial vehicle and landmark as the system function and the vision information as the observation. Complex additive noise model was adopted to describe the system noise in the filtering process. The wavelet-unscented Kalman filter (UKF) algorithm was obtained by introducing the wavelet analysis into UKF, thus the influence of vision observation noise on the filtering was inhibited successfully. Maximum a posterior (MAP) adaptive method was utilized to estimate the observation noise covariance matrix, which was further fed back into UKF to overcome the difficulties in identifying the covariance of observation after the wavelet de-noising. The simulation proved that the improvements in the filtering process to be effective in increasing the filtering accuracy.