Probabilistic tracking in a multitarget environment

This paper describes a new technique for probabilistic tracking in a multitarget, multi-sensor environment. The technique uses the doubly stochastic matrix as a basic structure for storing and manipulating data whose associations are ambiguous. A doubly stochastic matrix is a fuzzy mapping which represents the probabilistic associations between finite sets of objects in a balanced manner. Recursive operations on these matrices reflect the incorporation of new information into the system so that uncertainty is preserved for future resolution. In this way we avoid the use of statistical decision techniques which may introduce tracking errors by attempting to resolve ambiguity prematurely. Examples describing the basic operations are presented, and certain problems underlying the implementation of these ideas are shown to lead to the requirement for a more general approach.