Towards a non-contact driver-vehicle interface

A novel, non-contact, pointing interface is being developed for control of non-safety critical systems inside a vehicle with the aims of improving safety, decreasing manufacturing cost and improving the ease of driver migration between different cars. A driver operates the interface via an on-screen cursor using pointing gestures to be identified by a computer vision system. This paper describes the vision subsystem responsible for detection and tracking of the driver's hands. To be robust, it must detect and track under varying lighting conditions with no prior assumptions concerning the colour of the hands or clothing. Adaptive foreground and background models are used for segmentation and a robust geometrical hand model is employed for tracking. The system is demonstrated working at speeds close to real-time on a standard PC using image sequences captured inside a car.