Bohua Gan¹, Vincent Chang¹, Guanying Wang¹, Xiuli Pan¹, Guan Wang¹, Naihai Zou² and Felming Feng², ¹University of Michigan, China and ²Intel Asia-Pacific Research and Development Ltd, China

The paper introduces a case study of design and implementation of Intel-sponsored real-time face detection system conducted in University of Michigan—Shanghai Jiao Tong University Joint Institute (JI). This work is teamed up totally by 15 JI students and developed in three phases during 2013 and 2014. The system design of face detection is based on Intel High Definition (HD) 4000 graphics and OpenCL. With numerous techniques including the accelerated pipeline over CPU and GPU, image decomposition, two-dimensional (2D) task allocation, and the combination of Viola-Jones algorithm and continuously adaptive mean-shift (Camshift) algorithm, the speed reaches 32 fps for real-time multi-view face detection. Plus, the frontal view detection accuracy obtains 81% in Phase I and reaches 95% for multi-view detection, in Phase III. Furthermore, an innovative application called face-detection game controller (FDGC) is developed. At the time of this writing, the technology has been implemented in wearable devices and mobile with Intel cores.

Multi-view face detection, real-time, graphic processing unit (GPU), OpenCL, face-detection game controller (FDGC).