GPU-based Real-time Face Detector

The speed of face detection influences overall performance of applications like face recognition and human-robot interaction (HRI). In some applications such as long distance HRI, face detection is carried out on high resolution image up to 800x600 or more, leading to an increase in detection time and, in turn, deteriorating the application performance. To improve the speed of face detection, we propose a GPU-based face detection system consisting of modified census transform, AdaBoost and four phase cascades. To operate the face detection on GPU, we use CUDA. Experimental results demonstrated that the speed of our system than a CPU-based one was higher by a factor of five to eight compared with a CPU-based face detection system.