arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.00865cs.HC
收藏

基于商品WiFi的多用户活动识别与跟踪

Multi-User Activity Recognition and Tracking Using Commodity WiFi

Sheng Tan, Jie Yang

摘要:

本文介绍了MultiTrack,一种基于WiFi的人体感知系统,可以跟踪多个用户,并识别多个用户同时执行的活动。这样的系统可以实现用于多用户跟踪和感测的容易且大规模的部署,而不需要通过使用现有的WiFi设备(例如,台式机、笔记本电脑和智能设备)。其基本思想是借助多个WiFi链路和所有可用的5GHz WiFi信道,识别和提取与每个用户对应的信号反射。给定每个用户的提取信号反射,MultiTrack检查多个链路上的反射信号路径,以同时跟踪多个用户。它还重建每个用户的信号分布,就好像只有单个用户在环境中执行了活动,以促进多用户活动识别。我们在不同的多路径环境中评估MultiTrack,最多4个用户用于多用户跟踪,最多3个用户用于活动识别。实验结果表明,我们的系统可以达到分米定位精度和超过92%的活动识别准确率在多用户场景下。

英文摘要:

This paper presents MultiTrack, a commodity WiFi-based human sensing system that can track multiple users and recognize the activities of multiple users performing them simultaneously. Such a system can enable easy and large-scale deployment for multi-user tracking and sensing without the need for additional sensors through the use of existing WiFi devices (e.g., desktops, laptops, and smart appliances). The basic idea is to identify and extract the signal reflection corresponding to each individual user with the help of multiple WiFi links and all the available WiFi channels at 5GHz. Given the extracted signal reflection of each user, MultiTrack examines the path of the reflected signals at multiple links to simultaneously track multiple users. It further reconstructs the signal profile of each user as if only a single user has performed activity in the environment to facilitate multi-user activity recognition. We evaluate MultiTrack in different multipath environments with up to 4 users for multi-user tracking and up to 3 users for activity recognition. Experimental results show that our system can achieve decimeter localization accuracy and over 92% activity recognition accuracy under multi-user scenarios.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑