Traditional multi-target tracking methods usually divide the multi-target tracking problem into sub-problems such as track initiation, track maintenance and termination, data association, and single-target state filtering for study, without establishing a unified mathematical theoretical foundation. There is a clear boundary between data association and state filtering. State filtering depends on the results of data association. There are various combinations of data associations. When the number of targets and observations are large, it inevitably leads to the problem of combinatorial explosion.