一种基于压缩域的实时跟踪改进算法

An Improved Real-time Tracking Algorithm Based on Compressed Domain

在基于压缩域的实时跟踪算法中,判别函数对目标外观考虑不足易造成跟踪精度较低。为此,提出一种改进的基于压缩域的实时跟踪算法。利用稀疏测量矩阵提取候选目标的低维多尺度特征,并根据在线更新的特征概率分布,采用朴素贝叶斯分类器判别目标与背景,实现粗跟踪。通过视频帧间候选目标内部区域所具有的相似性,在粗跟踪的基础上实施基于动态目标外观模型的二次跟踪,在线寻找目标的最佳跟踪位置。对多种跟踪视频库的测试结果表明,该算法在不过量增加计算负荷的情况下能有效提高跟踪精度。

Aiming at the problem of low tracking precision caused by the discriminant function which has insufficient consideration for target appearance in the popular real-time tracking algorithm based on compressed domain, this paper proposes an improved algorithm. The low-dimensional multi-scale features of the candidate targets are extracted with a sparse measurement matrix. A Bayes classifier is adopted to discriminate the target and background according to online updating probability distribution of features, which realizes coarse tracking. On the basis of the coarse tracking result, the second tracking is carried out based on a dynamic appearance model of the target to search for the optimum tracking position online by measuring the local region similarity of the candidate targets between video frames. Test results for some challenging videos show that the proposed algorithm can improve the original tracking precision effectively without introducing too much computation.