Vision Control Strategy For Face Auto-Location And System Implementation

Face auto-location is one of the important subjects of applied vision research. Before locating a face in the image center, the system must detect the face location and get its feature point coordinate. Face auto-location belongs to the category of face tracking: it has been widely used in face recognition, human-computer interaction, intelligent video communication, surveillance and entertainment. Up to now, popular algorithms are Frame Change Detection, Motion Estimation and Shift-Coordinate. The first two algorithms need much information of correlative frames and have high computational requirements and the third one has poor precision. Therefore, although the technology is promising, it is not appropriate for system implementation.Herein, the idea, which is based on artificial neural network for the control strategy of face auto-location, is proposed. Neither requiring the information of correlative frames nor camera calibration, this algorithm can locate the feature point in the image center quickly. It can compute both the vertical and the horizontal rotation angles the pan-tilt camera needs to adjust when the coordinate of feature point in the image is required.The following is the whole process of face auto-location. To begin with, the paper presents a face detection and feature point acquisition method, which is based on KL skin-color, integration projection and small-gray-level-cluster. It reduces the image-processing region gradually, and has fast detection speed. Secondly, proposes a control strategy based on artificial neural network to control the pan-tilt camera to locate the face in the image center. Finally, the system is realized by VC++ programming. Although the system aims at a front-view face auto-location in simple background, it expands well in other conditions.Traditional face detection combining with the new control strategy based on artificial neural network is applied to face auto-location in this paper. The system based on this idea works fine and has practical value.