Target perceivability for multi-frame multi-target tracking

This paper presents a novel weight assignment model for a rank constrained continuous formulation of multiframe multi-target data association problem. The new weight assignment model explicitly estimates the joint probability of the target perceivability, the target state and measurement, within a Bayesian framework with continuous update when new measurements are available. The knowledge on target perceivability state at every time step also provides a new pruning criterion for reducing the space of all possible track associations. Experiments on synthetic data under different operating conditions demonstrate the effectiveness of the proposed weight assignment model for multi-target tracking problem. It also presents good qualitative tracking performance on real-world video data.