Ground target tracking and trajectory prediction by UAV using a single camera and 3D road geometry recovery

In this paper, we propose a new method to address the dual problem of ground target tracking in the image plane and 3D road geometry recovery using a single vision sensor on-board an unmanned aerial vehicle. We recover the road geometry from a single aerial image by a novel structure from motion algorithm with the simple assumption that road or lane boundaries are parallel curves and do not have significant twist. The coordinate of the ground target in the camera frame is then estimated using the recovered road geometry. An extended Kalman filter is finally applied to track and predict the trajectory of the target using recent target motion and the road geometry. The experimental results on simulated data and real world image show the feasibility of our approach.