Dynamic Probabilistic Models for Person Tracking and Activity Recognition

Object tracking and activity recognition underly many applications in computer science. Because they allow a principle integration of uncertainties, probabilistic models have been deepl investigated to solve object tracking and activity recognition tasks. In this talk, we present three probabilistic models we developed for: (i) head pose estimationvia probabilistic high dimensional regression; (ii) a variational Bayesian model for multi-persons tracking from cluttered visual observations, and (iii) an active speaker tracking based on a joint audio-visual probabilistic observation model.