GPU-BASED FACE TRACKING AT 500 FPS

Real-time face tracking at 500 fps was realized for an 8-bit 512x512 color image on a GPU-based high-speed vision platform by implementing a boosting-based face detection algorithm. This algorithm was developed by enhancing the Viola-Jones face detector. In the implementation, the size and position of a face pattern in an image can be efficiently extracted at high speed by reducing the number of window searches for Haar-like features and by combining skin color extraction with Haar classifiers, while assuming a property of high-frame-rate (HFR) vision, i.e., a small interframe image displacement. Several experiments were performed by using rapidly moving face patterns in order to verify the effectiveness of the proposed face tracking system.