Random Infinite Tree and Dependent Poisson Diffusion Process for Nonparametric Bayesian Modeling in Multiple Object Tracking

Abstract

Recent methods for tracking multiple objects have addressed important issues such as time-varying cardinality, unordered sets of measurements, and object labeling. Another challenge is how to robustly associate objects on a new scene with previously estimated objects. We propose a new method to track a dynamically varying number of objects using information from previously tracked ones. Our method is based on nonparametric Bayesian modeling using diffusion processes and random trees. We use simulations to demonstrate the performance of the proposed algorithm and compare it to a labeled multi-Bernoulli filter based tracker.