A Unified Joint Probabilistic Data Association with Multiple Models

Abstract : This paper presents the theory and examples of performance for a new algorithm that tracks targets using a Multiple Model Unified Joint Probabilistic Data Association (MM-UJPDA) filter. The models in the MM-UJPDA can be set to the ambiguity velocities encountered when initiating tracks on a sensor that has ambiguous velocities in its measurements. Alternately, the models can be set for tracking manoeuvring targets. Thus each parallel filter in the MM-UJPDAF is assigned one of a range of possible target model parameters. The term unified' summarizes a number of key features in the algorithm. These are: multiple non-uniform clutter regions, a model for a visible target to compute track confidence for track promotion, and measurement selection based on a fixed number of nearest measurements. The filter formulation used a new approach to create track clusters for determining the nearby tracks that share measurements. The filters performance is demonstrated with track initiation using the multiple model and multiple target approach while for established tracking only the multiple tar et a roach is used.