Multi-PN-learning for tracking applications
Jasper van de Ven, Arne Kreutzmann, Sascha Schrader, Frank Dylla · 2014
We present a general multi-target tracker able to simultaneously track, learn, and distinguish arbitrary objects in a single video stream and recognize them again after a temporal disappearance (reentering). We show how this tracker can be created as an extension of a general single-target tracker. Furthermore, we provide evidence that dissimilarities of tracked objects and relations in the learned knowledge can be exploited to improve individual tracking results.