视觉目标跟踪在视频智能监控和机器人视觉导航等领域有着广泛的应用。针对传统的核相关滤波器（KCF）跟踪方法缺乏目标尺度估计，对目标尺度显著变化的视频序列表现效果很差的问题，提出了一种多尺度估计的改进方法。该方法借鉴了分类尺度空间跟踪器（DSST）采用尺度金字塔相关滤波器做尺度估计的手段，将灰度图像金字塔映射到一维特征向量，然后用该向量作为尺度相关滤波器的输入，得到最高响应值后估计出目标尺度。使用benchmark数据集对改进的算法做了测试实验，并和其他已有的若干视觉跟踪方法进行了对比实验，验证了本算法的高效性。本算法在目标尺度变化、光照变化、姿态变化、部分遮挡、旋转及快速运动等复杂情况下均有较强的适应性。

Visual object tracking is widely used in various fields such as video intelligent monitoring and visual navigation of robot. Aiming at the problem that traditional kernelized correlation filter (KCF) tracking method is lack of target scale estimation, and performs poor in video sequence whose target scale changes observably, a multi-scale estimation modified method is proposed. This method uses reference for discriminative spacial scale tracker (DSST) to make scale estimation by adopting the scale pyramid correlation filter. The gray image pyramid is mapped to a one dimensional feature vector, and then this vector is used as input of scale correlation filter. The target scale is estimated from the highest response values. This modified method is tested on benchmark data set, and it is compared with the other current visual tracking methods. The results verify the high efficiency of the proposed algorithm. The proposed algorithm has strong adaptability under complex conditions, such as object scale change, illumination variation, posture change, partial sheltering, rotating and rapid movement.