Using optical flow for tracking

Abstract

We present two observation models based on optical flow information to track objects using particle filter algorithms. Although, in principle, the optical flow information enables us to know the displacement of the objects present in a scene, it cannot be used directly to displace a model since flow estimation techniques lack the necessary precision. We will define instead two observation models for using into probabilistic tracking algorithms: the first uses an optical flow estimation computed previously, and the second is based directly on correlation techniques over two consecutive frames. © Springer-Verlag Berlin Heidelberg 2003.