Abstract: Aiming at the problem of image target tracking, in order to improve tracking accuracy, an adaptive correlation filter tracking algorithm based on multi feature fusion is proposed. The algorithm first selects two complementary features of HOG and CN, and trains two correlation filter trackers to track image targets respectively. Then, the response map weight of two trackers is calculated by using the reliability formula proposed in this paper and the decision is made by adaptive fusion. In the update stage of the filter, the algorithm adjusts the learning rate of the filter dynamically by combining the response graph confidence of the two features and the rate of change between the two frames. The simulation experiment uses 36 groups of color video sequences in the tracking datum database （OTB-2013） to test and compare the popular correlation filter tracking algorithm. The results show that the algorithm is better than the other algorithms in the average tracking accuracy, and has certain application value.

Key words: correlation filter, target tracking, feature fusion, adaptive learning rate

摘要： 针对图像目标跟踪问题，为提高跟踪精度，提出了一种多特征融合的自适应相关滤波跟踪算法。算法首先选取HOG和CN两种互补特征，分别训练两个相关滤波跟踪器跟踪图像目标，然后利用提出的响应图置信度计算公式计算两个跟踪器的响应图权重并进行自适应融合做出决策。滤波器更新阶段，算法结合两个特征的响应图置信度与两帧之间的变化率动态调整滤波器学习速率。仿真实验采用跟踪基准数据库（OTB-2013）中的36组彩色视频序列进行实验，对比了流行的相关滤波跟踪算法，结果表明，该算法在平均跟踪精度上优于其他算法，具有一定的应用价值。

关键词: 相关滤波, 目标跟踪, 特征融合, 学习率自适应