Computer Science · Computer Vision and Pattern Recognition
Illumination Aware MCMC Particle Filter for Long-term Outdoor Multi-Object Simultaneous Tracking and Classification
Thierry Chateau, Datta Ramadasan, Universite ́ Blaise Pascal
See on the mapAbstract
This paper addresses real-time automatic visual track-ing, labeling and classification of a variable number of objects such as pedestrians or/and vehicles, under time-varying illumination conditions. The illumination and multi-object configuration are jointly tracked through a Markov Chain Monte-Carlo Particle Filter (MCMC PF). The measurement is provided by a static camera, associ-ated to a basic foreground / background segmentation. As a first contribution, we propose in this paper to jointly track the light source within the Particle Filter, considering it as an additionnal object. Illumination-dependant shadows cast by objects are modeled and treated as foreground, thus avoiding the difficult task of shadow segmentation. As a second contribution, we estimate object category as a ran-dom variable also tracked within the Particle Filter, thus unifying object tracking and classification into a single pro-cess. Real time tracking results are shown and discussed on sequences involving various categories of users such as pedestrians, cars, light trucks and heavy trucks. 1.