针对红外目标分辨率低、对比度差、信噪比低、纹理信息缺失等特点, 提出一种融合多特征的红外目标跟踪算法。利用背景感知相关滤波器生成大量真实样本, 对红外目标提取HOG特征和运动特征, 通过线性求和方式进行特征融合, 更好地发挥各自特征优势, 实现对红外目标运动的精准跟踪。另外, 提出使用空间加权窗代替传统相关滤波器中的余弦窗, 可以更加突出目标的中心位置, 同时也能很好地抑制边缘效应。采用VOT-TIR 2016数据集对算法性能进行评估, 同时和15种流行算法进行比较。结果表明, 本文算法在精确度和成功率上的得分分别为0.751和0.697, 在精确度和成功率指标方面分别提高了8.8%和15.4%, 具有一定的研究价值。

An infrared target tracking algorithm with multi-features was proposed in consideration of low resolution, poor contrast, low signal-to-noise ratio and lack of texture information of infrared target. The background perceptual correlation filter was used to generate a large number of real samples, and the HOG feature and motion feature were extracted for the infrared target. The feature fusion was performed by linear interpolation, and the advantages of the respective feature were well utilized to achieve accurate tracking of the infrared target motion. In addition, it was proposed to adopt the spatial weighting window instead of the cosine window in the traditional correlation filter to highlight the center position of the target and suppressed the edge effect. The VOT-TIR 2016 dataset was utilized to evaluate algorithm performance in comparison it with 15 popular algorithms. Simulation results show that the algorithm's scores on accuracy and success rate are 0.751 and 0.697 respectively. Furthermore, it is 8.8% and 15.4% higher than the second-ranking algorithm, which shows that the proposed algorithm has certain research value.