Higher Order Multiple Object Tracking for Crowded Scenes

The lifted disjoint paths formulation is a natural model for multiple object tracking. This model is able to obtain state-of-the-art results but is NP-hard. We present an efficient approximate message passing solver for LDP and integrate it into a multiple object tracker, which scales to very large instances that come from long and crowded scenes. We achieve comparable or better performance than stateof-the-art methods on MOT15/16/17 benchmarks and comparable results on the MOT20 benchmark. This has been out of reach up to now for known LDP-solvers due to the problem size and complexity of MOT20.