Tracking with intent

This paper presents the novel theory for performing behaviour-based tracking using intentional priors. Motivated by our ultimate goal of anomaly detection, our approach is rooted in building better models of target behaviour. Our novel extension of the Kalman filter combines motion information with an intentional prior. We apply our 'Intentional Tracker' to a pedestrian surveillance and tracking problem, using head pose as the intentional prior. We perform a statistical analysis of pedestrian head pose behaviour and demonstrate tracking performance on a set of simulated and real pedestrian observations. We show that by using intentional priors our algorithm outperform a standard Kalman filter across a range of target trajectories.