Research On The Algorithm Of Human Face Detection And Tracking

People in the image are always the point of the image, but according to the Human Visual Properties, the people usually pay more attention to the human face area of the image, therefore the face related imagery processing technology has become an important research direction of the computer vision field. The face processing technology has wide application and great development prospect in the field of identification verification, the man-machine interaction interface, the intelligent vision monitoring, the code and transmission based on ROI (Region of Interest), and the videophone.The purpose of this paper is to do some research on human face tracking based on face detection. So we choose the face detection algorithm which is speedy and effective, can increase the accuracy on the foundation of real time tracking.This content can mainly be divided into three parts: the possible face area judgment, face detection and face tracking.We do motion detection by fame subtraction method, get the motion area. Then do research on the Hsu R L skin color detection algorithm, do skin color detection on the motion area, get the possible face area which can reduce the face detection area and speed up the detection.When we are detecting face in the possible face area, this thesis do a deep research on the Adaboost algorithm, regarding the Adaboost classic algorithm doesn't have good effect on the circumstance of side face, we analyze the characteristics of the five senses organs in perspective conditions, improve the traditional Harr-like features, get some perspective features which can be used to do side face detection, then we use Adaboost algorithm to create the new sorter union structure by the side face features and traditional Harr-like features together. Experiment proves, in the condition of side, using features put forward by this thesis had better detection effect than use traditional Harr-like features only.The thesis proposes a face template of a muti-scales face five senses organs template based on the mosaic image method, then propose some side face template according to perspective. We limit the position of Harr-like features and side face features by those template, unite features and form a strong sorter., which can speed up face detection and can be used to detect face together with the sorter union structure of Adaboost algorithm. The experiment proved, using the method of the thesis had good practicability in face detection.Track the detected face as an object, the thesis forecasts face movement area by Kalman filter, then makes use of a template matching method of judging the similarity of image by the Covariance, which has good practicability. Experiments proved, tracking face in the condition of the face detection method taken by the thesis, was fast and accurate.