Real-time face detection for visual privacy protection in video surveillance system

Abstract

Recent studies of an intelligent video surveillance system deals with the visual privacy protection through deidentifying the human features. In this thesis, we propose a real-time face detection system for visual privacy protection in a video surveillance system. The previous studies in this field have many limitations, which are conducted in well-controlled environment. Unlike previous studies, proposed face detection system can be applied to an uncontrolled environment such as a cluttered and dynamic background. The proposed face detection unifies several existing detection algorithms like the background subtraction, the frame differencing, and the skin color model. The proposed system is consists of two phases. In phase one, the proposed method finds the movements of the object using the frame differencing and then it detects faces using color model in the boundary where movements occur during the background learning. In phase two, the proposed system sets the ROI of an object using background subtraction with the frame differencing, and then it recognizes faces through the skin color model and human features after background modeling. The experimental results show that the detection rate is 80.66% and that the averaged detection time per frame is less than 41msec on a 3.2GHz Pentium IV system.