JPDA intensity filter for tracking multiple extended objects in clutter

Abstract

The JPDA filter for multiobject tracking assumes that the number of objects is given. The objects are assumed mutually independent, but different objects may have different state spaces. In this paper the objects are extended, meaning that they can produce more than one sensor measurement, or highlight. Object highlights are modeled as a Poisson point process in the measurement space conditioned on the object's state. A new JPDA intensity filter (JiFi) is derived. It estimates an intensity function for each object, where the intensity is the expected number of highlights per unit object state space. A significant advantage of JiFi is that it avoids the extensive association probability calculations that are needed for JPDA. It is derived using an analytic combinatoric method based on probability generating functionals. Examples are given.