特征在线更新与加权的压缩跟踪算法

摘要

在目标跟踪中,针对目标外观改变使得目标丢失的问题,本文提出了特征在线更新与加权的压缩跟踪(compressive tracking,CT)算法.首先基于压缩感知理论提取目标的矩形特征,根据每个特征对当前帧目标的分类效果判定其可靠性,及时更新不可靠特征;其次,实时增加可靠特征在分类器中的权重,从而突出可靠特征的重要性;最后将加权候选样本特征输入贝叶斯分类器,得到下一帧的目标位置.选取八组视频序列测试改进算法的效果,结果表明与传统的压缩跟踪,局部敏感直方图跟踪(locality sensitive histograms tracking,LSHT)及在线自适应增强(online AdaBoost,OAB)算法相比,改进算法取得了更好的跟踪结果,并且在目标外观改变时依然跟踪准确,平均帧速为39fps,满足实时性要求.%In the object tracking,object is often lost because of the object appearance changes.Thus,an improved com pressive tracking algorithm based on the online feature updating and weighting is proposed in this paper.Firstly,the rectangle features are extracted based on the compressed sensing theory.The reliability of each feature is determined according to their classification performances for object tracking in the current frame.Then unreliable features are updated in time.Secondly,the values of reliable features' weights are increased in real-time such that their importances can be emphasized.Fi-nally,these new weighted candidate features are inputted into the Beyesian classifier to distinguish the object from background in the next frame.Eight challenging video sequences are chosen to verify the performances of our proposed algorithm.Compared with traditional compressive tracking algorithm,locality sensitive histograms tracking algorithm and online AdaBoost,experimental results show that our algorithm achieves better tracking results and is robust for appearance changes.The frame rate is 39fps in average,which satisfies the requirement of real-time tracking.