摘要
为了解决自适应运动目标跟踪中常见的漂移现象,提出一种基于主动特征选择(AFS)的在线权重多实例运动目标跟踪算法,在原有多实例学习的框架中引入一种新的包概率模型——加权和模型,然后使用主动特征选取法来提取更高效的特征,以降低建立分类器模型的不确定性,并通过优化Fisher信息判别准则进行在线boosting特征选取。实验结果表明,该方法鲁棒性较好,可以有效解决漂移问题,并能实时地完成在线跟踪。在此研究基础上将考虑更有效的光照不变特征。
To solve the common problem of drifting phenomenon in adaptive object tracking, this paper proposed an online weighted multiple instance object tracking method via active feature selection. In the prime framework of multiple instance learning, it drew a new kind of bag probability model-weighted sum model, and then used an active feature selection approach to get more informative features to decrease the uncertainty of classification model. Otherwise, optimizing the Fisher information criterion to select features in an online boosting method. Experimental resuhs show that this method is more robust, can effectively solve the drifting problem, and track the target real-time online. More effcient illumingtion invariant features will be considered based on this reserach.
出处
《计算机应用研究》
CSCD
北大核心
2015年第11期3463-3466,3470,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61373055)
