An integrated Monte Carlo data association framework for multi-object tracking

We propose a sequential Monte Carlo data association algorithm based on a two-level computational framework for tracking varying number of interacting objects in dynamic scene. Firstly, we propose a hybrid measurements generation process to facilitate varying number problems, the process mixes target-oriented measurements provided by target dynamics prior model and data-oriented measurements based on discriminative model. Secondly, an improved Monte Carlo joint data association filter is used to combat the curse of dimension problem. Finally, the particle based belief propagation is used to facilitate interactions among objects. This framework integrates discriminative model learning, Monte Carlo joint data association filtering, and belief propagation algorithm, these methods are realized as different levels of approximation to an 'ideal' generative model of multiple visual targets tracking, and result in a novel sequential Monte Carlo data association algorithm. The algorithm is illustrated via tracking many pedestrians in a real video sequence.