The derivation of multiple extended object intensity filter based on nonhomogenous poisson process

Most of traditional multiple target tracking algorithms depend on the fundamental assumption that one target at most produces one measurement at each time. However, this assumption is not yet appropriate for the current multiple target tracking scenes due to the high resolution capabilities of modern sensors. Several measurements can be generated by one target at the same time because of the high resolution capabilities. Under these circumstances it is more reasonable to treat the multiple target tracking as the multiple extended object tracking. The multiple extended object intensity filter is derived based on nonhomogenous Poisson process. The whole derivation is done in the framework of Bayesian theory. The multiple extended-object intensity filter consists of intensity predicting step and intensity updating step. The intensity predictor is exactly derived by Markov transformation of target state. The intensity connector is approximately done by marginal probability density, under the assumption that the observation process of extended object is a nonhomogenous Poisson process. The derived intensity filter provides an alternative to estimate the multiple extended-object states in the form of set.