GPU acceleration of real time Viola-Jones face detection
Adrian Wong Yoong Wai, Shahirina Mohd Tahir, Yoong Choon Chang · 2015
Face detection is a stepping stone to all facial processing systems such as face recognition with the task of determining face region from the input frame for applications like surveillance and law enforcement. However, face detection is a computational expensive process and thus, with acceleration it can influence the performance of the system. The latest Graphics Processing Unit (GPU) technology via Compute Unified Device Architecture (CUDA) has proven its capability to accelerate computation intensive algorithms to improve overall system performance. Thus, in this paper, a GPU acceleration of frontal face detection system utilizing Viola-Jones algorithm based on the Adaptive Boosting (Adaboost) using OpenCV and CUDA is presented. Experiments results show that the proposed GPU acceleration of face detection is able to achieve a speed up of up to 18 times, as compared to the conventional Central Processing Unit (CPU) version algorithm and yet maintain its detection accuracy.