摘要：

针对计算机视觉领域的目标跟踪问题,提出一种基于排序支持向量机的多特征融合目标跟踪算法。利用排序支持向量机学习得到排序函数,提取2 种不同的图像特征分别构造分类器,使2 个排序支持向量机并行预测,分别计算2 个分类器的错误率,从而得到分类器权重完成融合。实验结果表明,与目前主流的跟踪算法相比,该算法的跟踪结果更准确,在复杂视频环境下也能对目标进行稳定跟踪,具有较强的鲁棒性。

关键词: 目标跟踪, 多特征融合, 排序学习, 分类器, 排序支持向量机, 鲁棒性

Abstract:

For the object tracking problems in computer vision,this paper proposes a tracking algorithm based on Ranking Support Vector Machine (RSVM) fused with multiple features. Firstly,RSVM is used to get rank function. Secondly,the RSVMs combined with the two different image features are learnt respectively,then the two RSVMs predict parallel. Finally,the two RSVMs are fused with the weights which are calculated by the error rates of two classifiers,then it constructs a more adaptive RSVM framework fused with multiple features. This algorithm fuses image features effectively,and gets accurate predictions using RSVM. Experimental results demonstrate that it outperforms several stateof-the-arts algorithms.

Key words: object tracking, multiple features fusion, rank learning, classifier, Ranking Support Vector Machine (RSVM), robustness

中图分类号: 

• TP18