An Adaptive Particle Filter Method for Tracking Multiple Interacting Targets

In this paper, we present a new adaptive particle filter method for tracking multiple interacting targets. We introduce a sampling strategy to consecutively sample particles and adapt their spreading according to current measurements. For multiple targets, we develop a new concept of guiding the particles by current measurements of adjacent targets. Our method does not increase the computational complexity. Experimental results for tracking of multiple vehicles demonstrate that the search space is explored more efficiently. Further, we show that our method improves the robustness of the tracker when the target motion abruptly changes or the target is totally occluded.