基于核相关滤波器的颜色自适应目标跟踪算法

针对Staple算法中梯度直方图（HOG）特征和颜色直方图特征的融合无法自适应达到最优化的问题，本文提出了一种颜色自适应的核相关滤波目标跟踪改进算法，即Stronger-Staple算法（简称STR-Staple）。首先，本文用目标似然函数分别求出目标和背景所占比例的颜色直方图，并用巴氏系数实时测量目标与背景的颜色直方图的相似度，实现每一帧图像的跟踪监测；其次，提出一种自适应的融合系数，将相似度与融合系数相关联，对每一帧的特征匹配相应的权重，实现算法的最优融合。最后，本文算法在OTB-13和OTB-15两个数据集上与当前比较流行的5种跟踪算法进行比较。实验结果表明，该算法在光照变化、尺度变化、遮挡、变形、背景杂波等情况下均有较高的鲁棒性，且其跟踪精度、成功率在OTB-13数据集中分别为0.889、0.880。在OTB-15数据集中分别为0.741、0.644。均优于其它几种算法。 

Color-adaptive object tracking algorithm based on kernel correlation filter

Aiming at the problem that the fusion of the gradient histogram (HOG) feature and the color histogram feature in the Staple algorithm cannot be adaptively optimized, the paper proposes an improved color adaptive kernel-correlation filtering object tracking algorithm, namely the Stronger-Staple algorithm ( STR-Staple for short). First, the paper uses the object likelihood function to obtain the color histograms of the object and the background, respectively, and uses the Bhattacharyya coefficient to measure the similarity of the color histogram of the object and the background in real time to achieve the tracking and monitoring of each frame of image. Second, An adaptive fusion coefficient is proposed, and the similarity and the fusion coefficient are associated to match the corresponding weights of the features of each frame to achieve the optimal fusion of the algorithm. Finally, the algorithm in this paper are compared with the five popular tracking algorithms on the two data sets OTB-13 and OTB-15. The experimental results show that the algorithm has high robustness under lighting changes, scale changes, occlusion, deformation, background clutter, etc., and its tracking accuracy and success rate are 0.889 and 0.880 in the OTB-13 data set, respectively. In the OTB-15 data set, they are 0.741 and 0.644.which are better than other algorithms.