Dynamic Face Detection And Tracking Algorithm Base On Video Sequence

With the rapid development of science and technology and human needs of daily life, facedetection and tracking technology have become the hot topic in image processing and computervision. Followed the face detection and tracking technology based on the video sequencedevelopment and it has been applied to all fields in production and life, such as video conference,intelligent monitoring, etc.The paper studies a large number of face detection and tracking based on video sequence indomestic and foreign, analysis the current development situation of detection and tracking. Thispaper focuses on face detection and tracking algorithm for further research and explorationwhich based on video sequence. In order to improve the speed and accuracy of face detectionand tracking as the main line, the paper designs and realizes a kind of method which can detectand track faces in video in an effective and rapid way. Specific work includes:Firstly, due to the influence of the environment and the camera, video may be exposed tolight uneven, noise and external conditions which interference can reduce the detectionperformance. So the paper lucubrates many image preprocessing methods, it processes videoframe image in frame, after illumination compensation, filtering and so on, the quality of imagesare significantly improved. Experiments prove processed image has very important significancein improving the face detection accuracy.Secondly, at the real time request of face detection and tracking in video sequence withcomplex background, the paper proposes to extract skin feature from background as a candidatearea firstly. Then using PCA feature extraction algorithm which based on two-dimensional linearsubspace to reduce dimension of candidate area in a certain degree, it effectively reduced thecandidate for the size of the area once again.Thirdly, the paper studies application of the Adaboost algorithm in face detection. Itanalyzes the realization of weak classifier, strong classifier and cascade classifier training inAdaboost algorithm. In order to reduce cost data, this paper put forward two methods: enlargingrectangular features moving step length and eliminate the edge pixel method, which realizeimprovement and optimization on Harr-like characteristic quantity. Through the fusion color features, PCA characteristics and Adaboost classificationalgorithm this paper design a face detection system. With the accurate position using organgradient characteristics on detected face, Camshift can realize face tracking. The paper does a lotof experiments on several commonly face library and captured video. A great of experimentresults show that the design for dynamic face detection and tracking based on video sequence hasgood real-time quality, higher detection accuracy and takes shorter time, its tracking efficiency isbetter.