Transtrack: Online meta-transfer learning and Otsu segmentation enabled wireless gesture tracking
作者：
Highlights：
• Individual diversity causes poor performance of the current gesture tracking systems when they were directly applied to new users.
• An online meta-transfer learning method to learning the individual characters with low data collection cost.
• A data augmentation method that leverages the redundant information to generate virtual instances at the premises of the accurate detection result of recursive Otsu segmentation.
• A datum-based data alignment strategy that breaks the limitation of available classifiers for recognition without distort the instance.
摘要
•Individual diversity causes poor performance of the current gesture tracking systems when they were directly applied to new users.•An online meta-transfer learning method to learning the individual characters with low data collection cost.•A data augmentation method that leverages the redundant information to generate virtual instances at the premises of the accurate detection result of recursive Otsu segmentation.•A datum-based data alignment strategy that breaks the limitation of available classifiers for recognition without distort the instance.
论文关键词：Individual diversity,Meta-transfer learning,Gesture tracking,Channel state information,Data alignment,Online learning
论文评审过程：Received 27 December 2020, Revised 30 June 2021, Accepted 3 July 2021, Available online 5 July 2021, Version of Record 3 August 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.108157
