Vehicle tracking by simultaneous detection and viewpoint estimation

Abstract

We address the problem of vehicle detection and tracking for traffic monitoring in Smart City applications. We introduce a novel approach for vehicle tracking by simultaneous detection and viewpoint estimation. An Extended Kalman Filter (EKF) is adapted to describe the vehicle's motion when not only the pose of the object is measured, but also its viewpoint with respect to the camera. Specifically, we enhance the motion model with observations of the vehicle viewpoint jointly extracted by the detection step. The approach is evaluated on a novel and challenging dataset with different video sequences recorded at urban environments, which is released with the paper. Our experimental validation confirms that the integration of an EKF with both detections and viewpoint estimations results beneficial. © 2013 Springer-Verlag.