Efficient particle filtering for tracking maneuvering objects

Accurate tracking of elite athletes for performance monitoring allows sports scientists to optimize training to gain a competitive edge. An important challenge in this application is that the maneuverability of the athletes is high and the traditional Kalman filter (KF) will not provide satisfactory tracking accuracy. Further, high update rates, of the order of tens of updates per second for each player, are often required and hence, the tracking algorithm considered should be computationally efficient. In this paper we propose a computationally efficient multiple model particle filter (MM-PF) algorithm for tracking maneuvering objects. It uses a Gaussian proposal density based on the unscented KF and a deterministic sampling technique and provides tracking accuracy similar to that of the augmented MM-PF, but with much lower computational cost. The performance of the proposed algorithm was verified using simulations and data collected in field trials. The trials were conducted with the Australian Institute of Sport using a localization system we have designed.