摘要
复杂背景下实时目标的跟踪与识别属于自动目标识别(ATR)研究领域,包括对目标的分割、特征提取和目标识别等几个方面。由于现在的目标跟踪方法都是面向特定应用环境的,所以不存在一个算法能通用所有的场景。探索并明确算法的特点和应用环境,对于在实际应用中选择合适的方法是十分必要的。目前的大部分文章都是根据具体适应场景分析各自的方法,缺乏对跟踪方法的系统性研究,该文简要介绍了动态场景下单个运动目标的几种典型跟踪方法,在算法内容、假设条件、先验知识、理论计算量、实现难点及改进措施等方面进行了分析,并对研究难点及未来的发展趋势作了较为详细的阐述。
Real - time dynamic object tracking in moving background is a study field of Automatic Target Recognition （ATR）which includes target segmentation, character extraction and target recognition. Because the existing methods for object tracking are always aimed at special condition,there is not any algorithm which is universal. So it is necessary to explore and define algorithm＇s characteristic and applied condition for selecting a suitable method in practical application. So far,the majority of articles always analyze methods independently in specifical condition and it is lacking systemic study in object tracking. In this paper, some representative methods for single object tracking with background motion are introduced, and following major issues ： arithmetic content, hypothesis qualification, transcendental knowledge, calculating quantity, realizing difficulties and improved idea are analyzed. At the end, some detailed discussions on research challenges and future directions are also provided.
出处
《计算机仿真》
CSCD
2006年第5期181-184,共4页
Computer Simulation
关键词
目标跟踪
相关匹配
光流
帧差法
轮廓跟踪
Object tracking
Correlation matching
Optical flow
Frame difference
Contour- based tracking
