A multiple hypothesis walking person tracker with switched dynamic model
Geoffrey R. Taylor, Lindsay Kleeman · 2004
This paper presents the first comprehensive model for a walking person in range data from a scanner mounted at leg height. Our model helps to distinguish people from other types of moving targets and provides good tracking robustness. The central assumption of the walking model is that at least one leg always remains stationary. The current implementation tracks a single person in a multiple hypothesis framework. We extend the multiple hypothesis framework to allow for both association uncertainty and a switched dynamic model depending on the currently moving leg. Furthermore, an occlusion model and non-stationary dynamic state transition probabilities are used in the evaluation of hypotheses to further improve tracking robustness. Experimental results demonstrate the robustness and efficiency of the proposed framework. 1