Robust Real Time Face Tracking for the Analysis of Human Behaviour

We present a real-time system for face detection, tracking and characterisation from omni-directional video. Viola-Jones is used as a basis for face detection, then various filters are applied to eliminate false positives. Gaps between two detection of a face by the Viola-Jones algorithms are filled using a colour-based tracking. This system reliably detects more than 97% of the faces across several one-hour videos of unconstrained meetings, both indoor and outdoor, while keeping a very low false-positive rate (<0.05%) and without changes in parameters. Diverse measurements such as head motion and body activity are extracted to provide input to further research on human behaviour and for tracking participant activites at round-table meetings and similar discourse environments.