This paper presents a novel three-dimensional (3D) method for detecting, tracking and recognising human faces using a time-of-flight camera. The system works by detecting a single central feature point, typically the nose tip, and by intersecting the 3D point data with spheres centred at the central feature point. The resulting spherical intersection profiles are used to perform face recognition and to track the position and orientation of the face. The main benefit of this method is that it is fast and efficient in terms or memory and computational expense. Furthermore, as the system utilises a time-of-flight camera and topographical information, it is not affected by variations in illumination, face orientation or partial occlusion of facial features. Experimental results are provided which show the potential of this method to exceed the real-time performance of existing head-pose tracking and face recognition systems.
ANZSRC / FoR Code
080104 Computer Vision