视频序列对称差分法检测与预测人脸技术.pdfVIP

第 3 1 卷第 7 期 重 庆 大 学 学 报 Vol . 3 1 No . 7 　2008 年 7 月 J our nal of Chongqing U niver sit y J ul . 2008 　 ( ) 　　文章编号 :1000582 X 2008 视频序列对称差分法检测与预测人脸技术 印　勇 ,刘西富 (重庆大学 通信工程学院 ,重庆 400030) 摘 　要 :根据视频监控图像在时间和空间上的连续性和相关性 ,利用视频图像对称差分 ,找到 运动区域 ,再结合人脸肤色的聚类特征确定出人脸候选区域 ,改进了基于规则的人脸定位方法 ,利 用人脸的几何特征 ,实现复杂视频图像中的多人脸检测 。运用运动系数加上横向和纵向调节因子 , 对后续帧中的人脸加以预测 。实验表明 ,该算法复杂度小 ,准确率较高 ,对姿态 、表情 、背景等变化 情况下人脸的检测均具有较好的鲁棒性 ,预测跟踪效果好 。 关键词 :人脸检测 ;人脸跟踪 ;视频监控 ;对称差分 　　中图分类号 : TP39 1 文献标志码 :A Face detection and prediction technology based on video surveillance Yin Yong , Liu Xifu (College of Co mmunication Engineering , Chongqing U niver sit y , Chongqing 400030 , P . R . China) Abstract : Based on t he continuity and relativity in time and sp ace of video surveillance sequences , similar face areas were found in motion areas using symmetrical frame differences of video sequences and t he clu stering of skincolor features. By imp roving t he algorithm for face locating and u sing geometric facial features , multiple faces were detected in complex video sequences. The faces in video sequences were p redicted via motion quotiety and horizontal and vertical adj u stment factor s. The exp erimental result s show t hat t his algorithm has lower complexity , higher veracity , and better robu st ness in face detection in situations with changing gestures , exp ressions and backgrounds. The p rediction and tracking effect s are satisfied. Key words : f ace det ection ; f ace t rackin g ; video surveillance ; symmet rical f rame difference 　　随着科学技术的发展和人的安全意识的提高,