Tanker pose prediction based on modified unscented Kalman filter

Aiming at the low accuracy problem of the tanker pose tracking in the autonomous aerial refueling, a modified unscented Kalman filter(UKF) algorithm was put forward. The mathematical model of vision navigation was established. The Harris method was applied for corner detection, and then the random sample consensus( RANSAC) was used to match the detected corner. The historical forecast data was introduced into the current prediction so that the UKF prediction results could be modified with respect to the observation results from the corner match. As a result, the goal of high accuracy prediction was achieved. Compared with the standard UKF, the experimental results show that the prediction error of the proposed method is smaller than 5.8% which is feasible and effective in aerial refueling. The algorithm can eliminate the prediction error caused by the strong inference so as to effectively suppress the strong interference.