基于视频序列的面部特征跟踪系统的实现  

Framework for video-based facial features tracking and detecting

作　　者：池强[1] 朱永华[2] 方昱春[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200072 [2]上海大学计算中心,上海200444

出　　处：《计算机工程与设计》2011年第11期3819-3823,共5页Computer Engineering and Design

基　　金：上海市自然科学基金项目(08ZR1408200);上海市重点学科建设基金项目(J50103);上海大学国家大学生创新性实验计划基金项目(CXGJ09-10)

摘　　要：针对视频中人脸面部特征跟踪难以满足实时性与准确性要求的问题,提出了一种视频序列的面部特征跟踪系统。该系统利用视频流序列存在帧间相关信息的特点,进行面部区域粗定位;提出了一种Adaboost特征分类器训练方法,并使用该方法预先训练完成面部特征三元组(左眼,右眼,嘴部)的分类器进一步跟踪面部特征;最后提出了一个面部特征几何模型(facial feature geometrical model,FFGM),系统结合该模型仲裁检测的结果,最终实现了视频中面部特征的跟踪。实验数据的比较结果验证了该面部特征跟踪系统的可行性、实时性和准确性。To meet the two challenges that real-time and reliability are difficult for the tracking of facial features in video sequences, a video-based facial features tracking and detecting framework is proposed. The proposed framework utilizes the temporary related information between video sequences to reduce the region of facial features to be detected on. To further track the facial features, an improved adaboost cascade learning method is presented to find facial features by the triple-feature （left eye, right eye, mouth） cascades trained in advance. At last, a facial feature geometrical model （FFGM） is presented to arbitrate the detection result. Then facial features will be tracking successfully on the framework. Experimental results show that the developed framework is a robust and real-time solution for frontal facial features tracking.

关 键 词：面部特征 跟踪 ADABOOST 几何模型 实时 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]