Cascade particle filter for human tracking with multiple and heterogeneous cameras
Keisuke Kobayashi, Tamio Arai · 2009
In this paper, we propose a stochastic method for human tracking with heterogeneous cameras. Our tracking system employs two kinds of cameras, a foveated wide-angle lens and three network cameras. The tracking algorithm is based on cascade particle filter (CPF) involving two different weightings of particles. At the first stage of CPF, each particle is weighted by ground plane occupancy. At the second stage of CPF, each particle is weighted by similarity between color histograms. Each weighting utilizes an imaging-feature of each camera. Experimental results confirmed that the proposed method succeeded in human tracking.