Video-Based Face Detection And Recognition At A Distance

With the fast development of face detection and recognition and the practical requirements of video surveillance, video-based face detection and recognition at a distance has become an active research topic recently. Due to the uncontrolled illumination, the cluttered background and the long distance between camera and faces, video surveillance at a distance is quite complicated using face detection and recognition. As the size of the face region normally is very small, the resolution is quite low and the illumination changes unpredictably, it is a big challenge for us to deal with. In this thesis, we will do some researches on this topic. The main contents are shown in following:First, a human face detection method at a distance with various illuminations is presented in this paper. It employs a mixed multi-color model for face detection with low-resolution images, and then uses face hallucination method to reconstruct a high-resolution facial image which could be used for face recognition.Second, this paper proposes a feature extraction method based on LBP operator and Gabor faces, using histogram sequences to decrease the feature dimensions.Third, this paper applies the pseudo-inverse neural network classification method for real-time face recognition, and the result is satisfactory.Finally, a smart attendance system is designed and implemented based on face detection and recognition in a laboratory-oriented environment. It can record both entering and leaving without human intervention, which works in a completely automatic way.