CPHD filter addressing occlusions with pedestrians and vehicles tracking

In this paper, the problem of targets road tracking, like pedestrians and vehicles tracking is addressed. This paper proposes to improve a Cardinalized Probability Hypothesis Density (CPHD) filter in presence of occlusion using the sensor classification of each targets detected. Using this classification, a probability of target type is computed by Bayesian rules and used to deduce the width of targets. This width is necessary to take into account the occlusion problem in the Multi Target Tracking (MTT) filter. Besides, the probability of target type is also used to improve the performance of this MTT thanks to a new computation of the likelihood of measurements. Our system has been validated with real measurements from a smart camera in real traffic conditions.