核相关滤波（KCF）跟踪算法应用于尺度发生变化的视频跟踪时，因为无法跟上目标的尺度变化，极易受局部特征或者背景信息干扰，导致跟踪失败和目标丢失。提出了一种基于一维相关滤波的多尺度改进。该改进通过最小化损失函数求解相关滤波器用于尺度跟踪，并在跟踪过程中在线学习和更新滤波器。改进后的算法能够实时更新尺度信息，实现多尺度跟踪。多组实验结果和数据表明，改进后的算法在目标尺度发生剧烈变化时，相比KCF算法的中央位置误差降低了60%，距离精度和成功率分别提升了4.30%和28.65%，实现对目标持续稳定的跟踪。

When applied to visual tracking with scale change, Kernelized Correlation Filters tracking algorithm can’t follow the target’s scale change. The algorithm tends to get trapped by local features of the target or background information. This will lead to tracking failure. In this paper, a multi scale improvement based on one dimension correlation filter is proposed. A correlation filter for scale tracking is solved by minimizing the cost function. The filter is updated online through the tracking process. The scale information can be updated real time, and multi scale tracking is completed by the improved algorithm. Results and data of several experiments indicate that, when the target’s scale changes severely, the center location error is reduced 60% by the improved algorithm compared to KCF. The distance precision and success rate are reduced 4.30% and 28.65% respectively. The target can be tracked continuously and stably.