Design and implementation of efficient hardware solution based sub-window architecture of Haar classifiers for real-time detection of face biometrics

In this paper we propose a hardware solution by the use of FPGA based circuit for real time face detection. We have built a sub-window architecture for the extraction of Haar-like features, which are the basic elements of weak classifiers according to AdaBoost learning algorithm. The main contribution is that the proposed architecture removes traditional frame buffer, and only reserve the line buffer and sub-window register array. When the video data is in the progress of delivery, every pixel will be sent to the line buffer one by one and then be moved into sub-window for the circulation of weak classifiers calculation directly. Because integral image can be calculated by sub-window register array immediately, it provides the best detection performance without delay, and executes the minimal step of sub-window movement for best face detection accuracy. We have implemented hardware circuit and analyze it by Xilinx System Generator. The outcome shows that our design provides face detection speed up to 471.6 fps in the resolution of 640 × 480.