Estimation of Motion Parameters for Autonomous Land Vehicle from a Monocular Image Sequence

Abstract

This paper presents an approach to the estimation of the motion of a vehicle using a fixed order Extended Kalman Filter. The estimation assumes a fixed environment and a moving camera, mounted on the vehicle. The estimation of yaw rate and sideslip angle are shown to be possible using this technique. A feature correspondence based approach is used and the motion estimation algorithm employs the dynamics of the vehicle to increase the accuracy of the estimation. A simple algorithm to avoid the correspondence and occlusion problems is also developed. The algorithm depends on creating a sequence of virtual images from the real image sequence. This sequence has a fixed number of virtual feature points but retains the true motion information