基于多特征和尺度估计的KCF_MTSA算法

多模板尺度自适应核相关滤波器（KCF_MTSA）跟踪算法在目标移动模糊、旋转和尺度变化时跟踪距离精度与成功率较低。针对该问题，提出一种结合多特征和尺度估计的改进KCF_MTSA目标跟踪算法。采用方向梯度直方图和颜色名两种特征对目标进行表征，在训练阶段分别使用多模板核相关滤波器对上述特征进行训练，同时在检测阶段将两个滤波器的响应以权重形式进行自适应融合获取响应图实现目标定位，并使用一维相关滤波器进行目标尺度估计。实验结果表明，该算法的跟踪距离精度和准确率较改进前KCF_MTSA算法有明显提升，其距离精度和准确率分别提高15.8%和28.5%。

KCF_MTSA Algorithm Based on Multiple Features and Scale Estimation

The Kernelized Correlation Filter based on Multiple-Template Scale Adaptation(KCF_MTSA)tracking algorithm encounters a reduction in the tracking range precision and success rate when the target moves blurred,rotates and scales change rapidly.To address the problem,this paper proposes an improved KCF_MTSA target tracking algorithm combining multiple features and scale estimation.The Histogram of Oriented Gradient(HOG)and Color Name(CN)are used to represent the target.In the training phase,the multiple template Kernelized Correlation Filter(KCF)is used to train the above features.At the same time,in the detection phase,the responses of the two filters are adaptively fused in the form of weight to obtain the response map to realize the target localization,and the one-dimensional correlation filter is used to estimate the target scale.Experimental results show that the tracking range precision and accuracy of the proposed algorithm are significantly higher than the original KCF_MTSA algorithm,it improves the range precision and accuracy by 15.8%and 28.5%respectively.