基于帧差特征点的视频对象提取跟踪的研究  

Research on video stream object tracking using differential picture and point of interest

作　　者：韩建宇[1] 高胜法[1] 左凤艳[1] 

机构地区：[1]山东大学计算机科学与技术学院,山东济南250101

出　　处：《计算机工程与设计》2009年第24期5676-5680,共5页Computer Engineering and Design

基　　金：2008年度山东省自然科学基金项目(Z2008G03)

摘　　要：为了降低对象跟踪算法中特征点匹配的复杂度和无用特征点的数量,提高对象提取的精度以及跟踪的速度,提出了一种新的基于帧差特征点和边界点的对象提取、跟踪算法。首先结合形状信息以及自动阀值技术来减少特征点的堆积并提高特征点的利用率,其次从一个新的角度利用帧差技术,不仅有效的将大量的无关特点从跟踪系统中剔除,同时使跟踪范围缩小到一个更加合理高效的区域内。为了让物体的跟踪过程更加精确,通过添加边缘特征点来提高跟踪的鲁棒性。实验结果表明,该算法具有运算量小,精度高,可以处理对象遮挡跟踪问题并具有较强的跟踪鲁棒性的特点。To reduce the complexity of matching and the amount of useless interest points（IPs） during the tracking system and to enhance the precise of object extracting and object tracking,a new approach for extracting and tracking video stream object based on color interest points and differential pictures is proposed.Firstly,by making use of shape and automatic thresh holding to reduce accumu-lating and increase operation rate of IPs.Then through a new perspective of using differential pictures,not only large amount of useless IPs are eliminated but also the tracking areas is confined to reasonable bounds.In addition,the edge point added plays an important part in tracking accuracy and robustness.Less computation,more precise,robustness and have the ability to handle occluded objects are demonstrated by experiment on video sequences.

关 键 词：对象提取 对象跟踪 帧差 特征点 MPEG-4 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]