Using a genetic algorithm for multi-hypothesis tracking

Abstract

A technique has been devised that uses a genetic algorithm (GA) to address the multi-scan assignment problem in multitarget tracking. The problem is recast in the form of a scheduling problem, where the GA searches the space of possible orderings of detections, and a greedy heuristic is used to make the associations for a particular ordering. The resulting tracker can operate in either batch or continuous mode. In the continuous mode, a single population of hypotheses evolves on a fitness landscape that changes with each new scan of data.