摘要
目前,绝大多数智能眼镜都是基于触控板技术或语音控制实现人机交互,这些方法在驾驶环境或图书馆等场合存在限制.本文基于柔性压敏传感器设计了一个用户友好的免提式面部动作识别系统.该系统在眼镜脚的内侧集成一对柔性的压敏传感器,通过与面部皮肤贴合捕捉由面部动作引起的压敏信号.本文结合动态规划算法设计了一种连续信号分割算法来提取完整的面部动作信号.为了解决可穿戴设备领域普遍存在的弱样本问题,本文引入元学习机制并优化训练采样流程以提高模型的泛化性.结合真实数据集训练评估,系统识别准确率达到90%.面对用户个性化和处理弱样本的需求时,本文系统的性能表现出较高水平.
Currently,the majority of smart glasses rely on touchpad technology or voice control for human-computer interaction,which have limitations in scenarios such as driving or in quiet libraries.We present a user-friendly,hands-free facial action recognition system based on flexible pressure-sensitive sensors.This system integrates a pair of flexible pressure-sensitive sensors on the inner side of the glasses'temples,capturing pressure-sensitive signals induced by facial movements through contact with the skin.We employ a dynamic programming algorithm to design a continuous signal segmentation method for extracting complete facial action signals.To address the prevalent issue of weak samples in wearable devices,we incorporate a meta-learning mechanism into a convolutional neural network framework.Through training and evaluation with real datasets,the system achieves an accuracy rate of 90%.Our system outperforms several existing solutions in terms of personalized user experience and handling weak samples.
作者
张健
邢雄
ZHANG Jian;XING Xiong(School of Computer Science,Wuhan University,Wuhan 430072,China)
出处
《小型微型计算机系统》
北大核心
2025年第11期2625-2632,共8页
Journal of Chinese Computer Systems
基金
国家重点研发计划项目(2023YFB3106900)资助。
关键词
人机交互
可穿戴设备
面部表情识别
元学习
human-computerinteraction
wearable devices
facial activity detection
meta-learning
