Visual contour tracking based on particle filters

摘要

The Condensation algorithm, developed for visual tracking, is a variant of particle filter. In the sampling stage of Condensation, no use is made of the information from the current frame in the image sequence. As a consequence, the algorithm requires a large number of particles and is computationally expensive. In this paper, a Kalman particle filter (KPF) and an unscented particle filter (UPF) are applied to contour tracking to try to overcome the problem. These algorithms use the Kalman filter or unscented Kalman filter to incorporate information from the current frame. This sampling strategy can effectively steer the set of particles towards regions with high likelihood, and therefore can considerably reduce the number of particles needed for tracking. Performance comparisons show that the KPF is an improvement over Condensation, while the UPF has a much higher computational cost for equal tracking error.