Multi-Object Tracking Using a Generalized Multi-Object First-Order Moment Filter

The optimal approach to multisensor, multi-object fusion, detection, tracking, and identification is a suitable generalization of the recursive Bayes filter. Since this filter is computationally intractable in general, the first author has proposed an approximation of it based on propagation of a multi-object first-order moment statistic called the "probability hypothesis density" (PHD). Using more powerful proof techniques, we show that the original assumption of state-independent probability of detection can be removed. We also provide a less restrictive method for fusing multi-sensor data. A particle-systems implementation of the PHD filter is illustrated in a simple "toy" scenario.