基于字典学习的实时运动目标跟踪算法  

Robust Real-time Object Tracking Based on Dictionary Learning

作　　者：周安[1] 蒋辉[2] 余晋刚[1] 田金文[1] 

机构地区：[1]华中科技大学自动化学院,湖北430074 [2]北京机电工程研究所,北京100074

出　　处：《战术导弹技术》2014年第4期99-104,共6页Tactical Missile Technology

基　　金：国家自然科学基金资助项目(61273279)

摘　　要：采用提取图像的尺度不变特征可以获得较好的匹配跟踪效果,但该特征提取方法比较耗时。针对这一问题,提出了一种鲁棒的实时目标跟踪方法。该方法通过提取目标的多尺度平移、旋转特征来构建字典,提高了算法的鲁棒性。利用所构建的字典来表示待跟踪目标集特征,查找与目标模板最近邻的待跟踪目标,即可确定跟踪的最终结果。试验结果表明,这种基于字典学习的实时跟踪算法可以鲁棒实时地跟踪单目标。Through the matching of scale-invariant visual features, satisfactory tracking results can usual- ly be obtained. However, this approach computationally is very expensive. To overcome the challenging problem, a real-time object tracking method is presented based on feature matching. In the proposed ap- proach, multi-scale translation and rotation invariant features are extracted to build a dictionary with good robustness. And then, the target is represented by coding under this dictionary. Finally, the tracking re- sult is obtained by exhaustively searching for the one that best matches the target model among all the candidates. Experiment results show that this method can achieve real-time and robust tracking target.

关 键 词：目标跟踪 字典学习 实时跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]