A nonparametric Bayesian approach for enhanced pedestrian detection and foreground segmentation

With the continuous improvements in computer vision techniques, automatic low-cost video surveillance is becoming feasible. In the context of automatic surveillance, an important problem is the development of accurate models for foreground segmentation and pedestrians detection in outdoor scenes. In this paper we study an unsupervised algorithm based on infinite generalized Gaussian mixture models, that take into consideration the disadvantage of visible-light images (i.e. sensitivity to variations in illumination and lights) and infrared images (i.e. sensitivity to outdoor climate and temperature changes).