| This thesis is supported by the National 863 Plan project named New Method and Technology of Multi-pattern Biological Features Authentications Based on the Visual Sensation and the Cognition Mechanism.In recent years, research on face locating and tracking is one of the most popular topics in the image processing field. Along with the unceasingly thorough study, the research in this field has also made the considerable progress and has achieved very great achievement. The depth of study has been continuous increased, the scope has also been unceasing expanded. The research on human face locating and tracking has important academic value, great practical application value and market potential. So the request on this research has already become even more urgency. At present, the human face localization and tracking technology have already applied to various situations such as military, medical service, bank, customs security check. Moreover, along with the gradual maturity and consummation of this technology, it will apply to on each aspect of social and life, and provide powerful safeguard for people's stable life and steady development steps for harmonious society.This thesis is mainly consisted by two parts.The first part is the human face locating. In this part, two consecutive video images are pre-processed to remove the static parts in background, and to reduce the processing scope to the moving area, and then to unite face skin color model for face preliminary detection, after this step, the human face can be roughly located. However, in order to increase the locating the tracking accuracy, we take advantage of adaboost algorithm based on statistics for face precise detection, and label the face by an external rectangle. The innovation of this part is additional function of detecting rotating face in different angle on original adaboost algorithm. This algorithm judges the face's rotation degrees through the laplacian directive edge detection operator united the radiation template, and then changes the rotated face to the positive one in order to be detected by adaboost algorithm. In the second part, that is the human face tracking part, it take effective advantage of face detection results, and using the combination of adaboost and camshaft algorithm for face tracking. Initial the camshaft search windows by the size and location of the face external rectangle resulted by previous adaboost algorithm. It only needs to search the image during a bigger size than the rectangle while tracking the next frame.The algorithm of frame difference method,human face skin color model method combined with adaboost algorithm and the innovation method of additional function of detecting various angle of rotated face we took in the locating part of this paper are verified by experiments that they can greatly enhances the detection efficiency under the guarantor of face examination precise rate and has the very good examination effect to the rotated face. In the human face tracking part, it improves iterative process of the original camshift algorithm while initializing the search window. To initialize the search window with the locating results from the first part reduces the iterative times of camshaft algorithm, and raises the track efficiency greatly. |
