Fast Face Detection And Face Tracking

If the face detection wants to be practical applications, accuracy and speed are the two key issues that need to be resolved. After 10 years of development, the accuracy of face detection has been significantly improved, but the speed is still a problem to cumber face detection system from being widely used. Therefore, how to build fast and practical face detection and face tracking algorithms becomes the major problem of this paper.In this paper, we introduce color constancy algorithm into human skin segmentation to solve the problem of skin color segmentation under varying illumination. Firstly color constancy algorithm is adopted to estimate the illumination, then Gauss Skin Models and automatic-thresholding technique is used for skin-color segmentation in YCbCr color space. The experiment result shows that this method is good for skin segmentation in the image which is taken under deviant illumination and complex background. Applying this new method to our laboratory's face detection system, the original system's detection rate has increased by 5.4%.The Adaboost face detection algorithm presented by Viola contains three parts: Rapid calculation of Harr-like features; Using Adaboost algorithm to combine several weak classifiers into a strong classifier; Serializing several strong classifiers to constitute the final cascade classifier. After studying this method deeply and analyzing the edge characteristics of face, we proposed an improved algorithm-adding Edge direction Feature. The EDF features and the extended Harr-like features are combined to construct new feature library, then Adaboost algorithm is used to train classifier. The experimental results show that the proposed algorithm could reach a higher detection rate on smaller training sample set, and the improved algorithm reaches higher detection rate.Driven by application demand of Multimedia Teaching System, In this paper we combine Adaboost face detection algorithm and Camshift tracking algorithm, taking Adaboost algorithm as initial face location, and Camshift algorithm as follow-up frames tracking, to construct an effective and practical face tracking system which is very suitable for our Multimedia Teaching System. In simulation, the algorithm has achieved good results.