基于核相关滤波器的尺度自适应追踪算法

Scale adaptive tracking algorithm based on kernelized correlation filters

针对目标追踪算法里核相关滤波器(KCF)不能处理尺度变化的问题,设计了一种用于尺度估计的系统。在标准KCF获取的位置周围提取尺度样本,并将其与一个额外的核相关滤波器进行卷积,可将卷积结果中的最大值位置作为目标在新帧中的尺度大小。在OTB数据集中的结果显示出算法可以高速有效地应对目标追踪中的尺度变化问题。

Aiming at the problem that kernelized correlation filter(KCF)is unable to handle scale variation in object tracking algorithm, a system for scale estimation is designed.Around the position estimated by a standard KCF,another filter is convolved with the signal sampled around the position.The maxima in output is treated as a suitable scale in new frame.The experimental results on OTB dataset show that it is able to handle scale variation efficiently.