Pedestrian tracking using Random Finite Sets

For most multi-target tracking applications it is assumed that the movements of the objects are independent of each other. The validity of this assumption depends amongst others on the measurement rates of the sensors and the distance between the objects. In scenarios with a high object density the measurement rates for some of the objects may decrease due to short-time occlusions. Integrating the dependencies among the objects during occlusions should therefore improve the performance of the system. Within the finite set statistics (FISST) it is possible to model these dependencies and to integrate them into a Bayes filter. In this contribution a sequential Monte Carlo multi-target Bayes (SMC-MTB) filter based on FISST is used for pedestrian tracking. Furthermore, a model which avoids collisions of the pedestrians as well as a state dependent detection probability are integrated into the filter. The results of the SMC-MTB filter are evaluated using real sensor data and compared to the results of a CPHD filter.