基于特征鉴别性分析的变尺度核相关滤波跟踪算法

Variable scale kernel correlation filtering tracking based on feature discriminant analysis

针对核相关滤波跟踪算法对场景依赖及无法适应目标尺寸变化问题,提出了一种特征鉴别性选择分析的变尺度核相关滤波跟踪算法。在核相关滤波跟踪框架内,提取目标的颜色、纹理、梯度特征,建立样本集合,以最小均方损失函数设定各特征样本权重,鉴别性地选择出最优和次优的两种特征进行自适应融合。在此基础上,利用高斯金字塔构建一维尺度相关响应滤波器,对目标的尺度变化进行估计;通过主旁瓣均值比对融合后的跟踪结果进行评判,实现模板的差异化更新。理论分析和实验表明:所提算法在遮挡及光照变化场景具有较高的跟踪精度并对目标的尺度变化具有一定的估计能力。

Aiming at problem of scene dependent and can not adapt to the change of target size of kernel correlation filter tracking algorithm,propose a variable scale kernel correlation filtering tracking algorithm based on feature selection and analysis.Within the framework of filting tracking of kernel correlation extract color,texture,gradient feature of target,set up sample set.According to the minimum mean square loss function,set sample weight of each feature,differentially select optimal and suboptimal two kinds of features for adaptive fusion.Using Gaussian Pyramid to construct ID scale correlation response filter,to estimate scale variations,compare sidelobe mean value with tracking results fusion to realize difference update of the template of target.Theoretical analysis and experimental results show that the proposed algorithm has higher tracking presision and has certain estimation ability to scale variation of target.