融合多模态生成和情景训练的环境无关手势识别  

Environment-independent Gesture Recognition Based on Multi-modal Sample Generation and Episodic Training

作　　者：程宇 周瑞[1] 张子若 罗悦 张宏旺 王佳昊[1] CHENG Yu;ZHOU Rui;ZHANG Ziruo;LUO Yue;ZHANG Hongwang;WANG Jiahao(School of Information and Software Engineering,University of Electronic Science and Technology of China,Chengdu 610054,China)

机构地区：[1]电子科技大学信息与软件工程学院,成都610054

出　　处：《小型微型计算机系统》2024年第2期265-270,共6页Journal of Chinese Computer Systems

基　　金：电子科技大学-智小金-智能家居联合研究中心项目(H04W210180)资助;四川省科技支撑计划项目(2021YFG0024)资助。

摘　　要：无线通信和感知技术的发展促进了WiFi感知的产生与发展.依据人体及其行为对WiFi信号传播的影响,通过模式匹配可以实现基于WiFi的手势识别、活动识别、定位等感知任务.但是WiFi信号对环境具有较大依赖性,目标人员或周围环境的变化会导致已经建立的感知模型失效.为了解决这个问题,现有方案通常采用半监督或无监督域适应方法.但在实际应用中,无法预先获得新环境中的数据.因此,需要一种无需新环境数据,即可自动泛化到新环境的方法.为了实现这个目标,本文提出一种基于多模态样本生成和情景训练的环境无关手势识别方法.该方法采用若干源域的数据建立手势识别模型,能够在目标域没有任何数据的情况下,泛化到目标域中.实验结果表明,该方法在目标域无数据的情况下,对新用户和新环境的手势识别正确率均超过80%,高于业界现有水平.The development of wireless communication and sensing technology promotes the generation and development of WiFi sensing.According to the influence of human body and its behavior on WiFi signal transmission,WiFi-based gesture recognition,activity recognition,positioning and other sensing tasks can be realized through pattern matching.However,WiFi signals have a great dependence on the environment,and the changes of the people or the surrounding environment will lead to the failure of the established sensing model.To solve this problem,existing schemes usually adopt semi-supervised or unsupervised domain adaptation methods.But in practical applications,the data in the new environment cannot be obtained in advance.Therefore,we need a way to automatically generalize to new environments without the need for new environment data.To achieve this goal,this paper proposes an environment-independent gesture recognition method based on multi-modal sample generation and episodic training.This method uses the data of several source domains to build a gesture recognition model,which can be generalized to the target domain without any data.Experimental results show that the accuracy rate of gesture recognition for new users and new environment is more than 80%,which is higher than the existing level in the industry.This method is universal and can also be applied to other WiFi sensing tasks.

关 键 词：WiFi感知 手势识别 环境无关 虚拟样本生成 情景训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]