Abstract:

In complex scenes and drastic changes in target appearance, the tracking model of the Kernelized Correlation Filter（KCF） method is susceptible to interference, which results in the problem that the tracking window is not adaptive and the target is lost. To this end, a mobile robot tracking system based on depth information is proposed. The target scale frame is estimated by Cross-Searching Edge（CSE） method, and the tracking failure is checked by fluctuations in axial relative kinetic energy method. The target loss is recovered by scale pool and search strategy. Experimental results show that the proposed method combining KCF and scene depth information can effectively realize target tracking window self-adaptation and target recovery after losing, which has a stable application effect on mobile robots.

Key words: target tracking, depth information, Kernelized Correlation Filters（KCF）, mobile robot

摘要：

在复杂场景及目标外观剧烈变化条件下，核相关滤波器方法跟踪模型易受干扰，造成跟踪窗不能自适应及目标丢失问题，提出一种融合深度信息的移动机器人跟踪系统。通过边缘交叉搜索法进行目标尺度框估计，通过轴向相对动能波动法进行跟踪失败检查，利用尺度池和搜索策略实现目标丢失找回。实验结果表明，结合核相关滤波器和场景深度信息的方法可以有效实现目标跟踪窗口自适应、目标丢失后找回，在移动机器人上具有较稳定的应用效果。

关键词: 目标跟踪, 深度信息, 核相关滤波器, 移动机器人