The PDAF based Active Contour

We present a new active contour model in spatiovelocity space which is based on the probability data association lter (PDAF) approach. It employs a directional-based measurement model of image potential and of opticalow along the contour points. The proposed directional approach of measurements is the basis for the velocity-based discrimination between measurements of the object to that of image clutter. The model chooses the appropriate measurements which are most consistent with previous estimation of motion in the sense of the PDAF approach. In order to obtain reliable measurements of image motion and of gradientbased image potential, we propose a directional smoothing operator which is the basis for discriminating the objects measurements from that of image clutter. The method was applied to real world tracking problems such as a walking leg and a waving hand.