Tracking Targets under Uncertainty: Natural Computing Approaches

Tracking or more generally state estimation of dynamic systems are tasks that appear in many different contexts - for instance in surveillance with wireless sensor networks. Usually the state-evolution equations are assumed to be known excepting some parameters. In this case, particle filters and related approaches have been applied with great success. Very few attempts, however, have been made so far to address the problem of an unknown state equation. This paper presents approaches based on natural computing to solve this difficult and complex situation leading to a new kind of algorithms. Improvements to the original methods are introduced and investigated. The tracking quality is examined in simulations and compared to that of particle filters. The results show the performance of natural computing approaches are similar to that of particle filters for systems with known state-evolution equations. The new methods, however, can also be applied in situations with severe uncertainties.