A comparison between Unscented Kalman Filtering and particle filtering for RSSI-based tracking

The task of tracking targets carrying active radio-frequency identification (RFID) tags based on the received signal strength indication (RSSI) values of tag transmissions is a classical Bayesian filtering problem. Since the problem is nonlinear, no closed-form solution is known and tractable approximations must be used. Unscented Kalman Filtering (UKF) and Particle Filtering (PF) are two leading candidates proposed in literature. However, a head-to-head comparison of the two is currently unavailable. In this paper, we address this issue by comparing and contrasting these two tracking techniques in terms of their tracking accuracies and consistencies in various scenarios. Based on extensive simulation results as well as real-life experimental data, we conclude that the UKF significantly underperforms relative to the PF in two realistic scenarios: (i) when there are significant co-dependencies in the motion of the targets, and (ii) when a diverse radio environment affects the propagation characteristics of the tag transmissions (like occlusions, multipath and shadowing). The second situation is especially significant because it implies that the success of the UKF is contingent on a free-space like environment. Therefore, it is not a robust solution in practice.