A Branch-and-Bound Algorithm for Multiple-Target Tracking and its Parallel Implementation

Abstract

Multiple-target tracking is used to identify the traveling paths of multiple point targets from a set of detected measurements. Since all the detected measurements often have a uniform look, it becomes difficult to distinguish one target from another, and targets from false alarms. Based on the statistical information about channel noise, target initiation rate, false alarm rate and probability of detection, a multiple-hypothesis testing can be formulated to associate each measurement with a specific source. However, this association process is a computationally explosive problem. By converting the association problem to an equivalent assignment problem, a branch-and-bound algorithm can be applied to provide an efficient method for generating hypotheses, evaluating their likelihood, and identifying the leading N most likely hypotheses. The modularity of the branch-and-bound algorithm leads naturally to a parallel computer implementation using the best-first search strategy.