Exploiting superpixel and hybrid hash for kernel-based visual tracking
作者：
Highlights：
• Superpixel analysis approach is employed to reconstruct target appearance in visual tracking.
• Representative superpixel blocks combined with kernel-based filter are exploited for roughly target location.
• Superpixel block modification method based on hybrid hash is proposed to accurately locate target.
• Conventional kernel-based filter is improved with peak-response scale estimation and color features.
• Comprehensive experiments on publicly available datasets demonstrate our tracker superior to state-of-the-art trackers.
摘要
•Superpixel analysis approach is employed to reconstruct target appearance in visual tracking.•Representative superpixel blocks combined with kernel-based filter are exploited for roughly target location.•Superpixel block modification method based on hybrid hash is proposed to accurately locate target.•Conventional kernel-based filter is improved with peak-response scale estimation and color features.•Comprehensive experiments on publicly available datasets demonstrate our tracker superior to state-of-the-art trackers.
论文关键词：Object tracking,Kernel-based filter,Superpixel clustering,Hybrid hash analysis
论文评审过程：Received 13 June 2016, Revised 9 February 2017, Accepted 8 March 2017, Available online 9 March 2017, Version of Record 17 March 2017.
论文官网地址：https://doi.org/10.1016/j.patcog.2017.03.015
