Multi-Camera tracking with a Probabilistic Occupancy Grid

Опубликовано 6 сентября 2016, 6:29

Joint work with Jérôme Berclaz, Richard Lengagne and Pascal Fua. The topic of this talk is the description of an original multi-camera tracking algorithm which combines a fine frame-by-frame stochastic estimate of the ground occupancy with a Hidden Markov Model. Given three or four synchronized videos taken at eye level and from different angles, it follows up to six individuals across thousands of frames in spite of significant occlusions. Our contribution is twofold. First, we demonstrate that we can handle occlusions at each time frame independently, even when the only data available comes from the output of a simple blob detector based on background subtraction and when the number of individuals is unknown a priori. This is achieved by computing the marginals of a product law minimizing the Kullback Leibler divergence to the true posterior given the background subtraction result. Second, we show that multi-person tracking can be reliably achieved by computing most probable trajectories separately over long sequences, provided that a reasonable heuristic is used to rank these individuals and avoid confusing them with one another.