Unsupervised learning-based long-term superpixel tracking
作者：
Highlights：
• A video processing pipeline dedicated to long-term superpixel tracking is proposed.
• Multi-step superpixel pairings are estimated in an unsupervised learning fashion.
• Unsupervised learning exploits robust context-rich features extended to multi-channel.
• Elementary matches are combined through multi-step integration and majority voting.
• Accurate long-term superpixel matches are shown for video object tracking experiments.
摘要
•A video processing pipeline dedicated to long-term superpixel tracking is proposed.•Multi-step superpixel pairings are estimated in an unsupervised learning fashion.•Unsupervised learning exploits robust context-rich features extended to multi-channel.•Elementary matches are combined through multi-step integration and majority voting.•Accurate long-term superpixel matches are shown for video object tracking experiments.
论文关键词：Superpixel matching,Unsupervised learning,Superpixel tracking,Multi-step integration,Random forests,Forward-backward consistency
论文评审过程：Received 26 January 2018, Revised 14 June 2019, Accepted 27 June 2019, Available online 19 July 2019, Version of Record 27 August 2019.
论文官网地址：https://doi.org/10.1016/j.imavis.2019.06.011
