Research Of The Real-time Face Detection Based On Video

Face detection is the most active and challenging tasks for computer vision and pattern recognition. It can be widely applied to such fields as video surveillance, human computer interface, content-based image retrieve, teleconference, personal identification, virtual reality, etc. With the rapid development of intelligent computing technology, new methods and new technologies will continue to be used, which bring more vitality to the research.In this thesis, after analyzing the face detection methods currently used by others, we present the multi-views face detection method based on improved AdaBoost and multiple decision tree. The main work is described as bellows:(1)According to the different angle of multi-view faces and automatic multi-view faces samples space partition, a multi-view face detection method based on multiple decision tree is proposed. The FCM algorithm in data mining field is introduced for automatic multi-view faces samples space partition, which solves the problem of Multi-pose face uncertainty angle. The sample space splits under the principle, which guarantees that the same detection rate is achieved by using Calculation at least.(2)After delving into problems in AdaBoost algorithm such as haar-like features, cascade classifier, detect results processing, etc. The new image traversal method using adaptive step length is presented to avoid repetition of faces detection operations. The method uses an inverted pyramid structure on images traverse and use the mirror image of the map control testing sliding window. the method takes advantage of the cascade classifiers' characteristics of high rate detection , avoids repetition of faces detection operations, reduces computing time and improves efficiency of detection.(3)By using of the human faces' movement features in video, auto-tracking the results of face detection method is proposed based on the existing target tracking algorithm. The method avoids detecting every frame of video. It can save computing time, improve real-time performance of system. By the characteristics of human faces in video, the tracking objects process is determined to failure or not. If the tracking process failures, the faces detection will run automatically.(4)In order to verify the effectiveness and feasibility of the improved algorithm, a number of experiments based on test data sets are carried out. Those experimental results show that: the multi-view faces can be detected with our improved face detection system based on Adaboost algorithm, and the detection accuracy rate is improved than the existing detection system.