Head-pose tracking with a time-of-flight camera

Intelligent interfaces that make use of the user's head pose or facial features in order to interpret the user's identity or point of attention, are finding increasing application in numerous fields. Although various techniques exist to passively track the user's gaze or head pose using monocular or stereo cameras, these systems generally cannot perceive in detail the characteristic three-dimensional (3D) profile of the user's head or face. Time-of-flight cameras, such as the Swiss Ranger SR-3000, are a recent innovation capable of providing three-dimensional image data from a single sensor. The advent of such sensors opens up new possibilities in the fields of head-pose tracking and face recognition. In this paper we propose a novel 3D time-of-flight camera system for robust and accurate head-pose tracking in real-time. Our system requires no manual calibration and is tolerant of varying lighting conditions, occlusion and facial expressions. The system can also be used to generate accurate 3D "face prints" suitable for use in face recognition with minimal data and search times. Preliminary experimental results are provided which demonstrate the potential of the system.