Multi-object tracking via a recursive generalized likelihood approach

This paper deals with the problem of tracking multiple objects with multiple sensors where the association of measurements with objects is ambiguouus. Object motion is modeled as a random process moving locally about a mean path where the random process model can be one of a discrete set of possibilities. In the above setting, the tracking problem amounts to associating data with objects, selecting motion models for the objects and estimating the object state. The multi-object tracking problem is solved using the generalized likelihood approach. No a priori statistical information is used concerning the correctness of a data association hypothesis. A practical recursive algorithm is described that has been successfully applied to large scale surveillance problems.