Particle filter to track multiple people for visual surveillance

A particle filter (PF) has been recently proposed to detect and track colour objects in video. This study presents an adaptation of the PF to track people in surveillance video. Detection is based on automated background modelling rather than a manually generated object colour model. Furthermore, a labelling method is proposed to create tracks of objects through the scene, rather than unconnected detections. A methodical comparison between the new PF tracking method and two other multi-object trackers is presented on the PETS 2004 benchmark data set. The PF tracker gives significantly fewer false alarms owing to explicit modelling of the object birth and death processes, while maintaining a good detection rate.