Kinematic and Attribute Fusion Using a Bayesian Belief Network Framework

摘要

The focus of tracking applications has traditionally centred on kinematic state estimation. However, attribute information has the potential to not only provide identity and class information, but it may also improve data association and kinematic tracking performance, Bayesian Belief Networks provide a framework for specifying the dependencies between kinematic and attribute states. Algorithms based on this framework are developed for joint kinematic and attribute data association, kinematic tracking, attribute state estimation, and joint kinematic and attribute tracking. The algorithms are demonstrated using simulated tracking scenarios.