Abstract:

We present an object tracking framework that employs Dirichlet Process Mixture Models (DPMMs) in a multiple hypothesis tracker. DPMMs enable joint detection and tracking of an unknown and variable number of objects in a fully automatic fashion without any initial labeling. At each frame, we extract foreground super pixels and cluster them into objects by propagating clusters across consecutive frames. Since no constraint on the number of clusters is required, we can track multiple cluster hypotheses at the same time. By incorporating super pixels and an efficient pruning scheme, we keep the total number of hypotheses low and tractable. We refine object boundaries with Markov random fields and connectivity analysis of the tracked clusters. Finally, we group tracked hypotheses to combine possible parts of an object as one.