Rits Eye: a software-based system for real-time face detection and tracking using pan-tilt-zoom controllable camera

Abstract

This paper describes a working system that can robustly detect and track human faces with unknown sizes and positions in complex backgrounds from color images in real time. Skin color is modeled as and detected by a 3D Gaussian distribution in the RGB color space among different persons within the same race. Detected skin pixels are grouped into regions whose centroids and areas are recorded. If dark regions for eyes, eyebrows and mouth are also detected at certain positions within these regions, they are regarded as faces. Once the faces are detected, the system starts to track one of them using a pan-tilt-zoom controllable camera. Without any special hardware, the system achieves a speed of 0.2 sec for face detection and 0.1 sec for face tracking, on a Pentium 266 MHz PC, including the image acquisition time.