JOINT HUMAN ORIENTATION-ACTIVITY RECOGNITION USING WIFI SIGNALS FOR HUMAN-MACHINE INTERACTION Hojjat Salehinejad12 Member IEEE Navid Hasanzadeh2Radomir Djogo2 and

2025-05-05 0 0 604.56KB 5 页 10玖币

侵权投诉

JOINT HUMAN ORIENTATION-ACTIVITY RECOGNITION USING WIFI SIGNALS FOR

HUMAN-MACHINE INTERACTION

Hojjat Salehinejad1,2, Member, IEEE, Navid Hasanzadeh2,Radomir Djogo2, and

Shahrokh Valaee2, Fellow, IEEE

1Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, MN, USA

2Department of Electrical & Computer Engineering, University of Toronto, Toronto, Canada

hojjat@ieee.org, {navid.hasanzadeh, radomir.djogo}@mail.utoronto.ca, valaee@ece.utoronto.ca

ABSTRACT

WiFi sensing is an important part of the new WiFi 802.11bf

standard, which can detect motion and measure distances.

In recent years, some machine learning methods have been

proposed for human activity recognition from WiFi signals.

However, to the best of our knowledge, none of these meth-

ods have explored orientation prediction of the user using

WiFi signals. Orientation prediction is particularly critical for

human-machine interaction in an environment with multiple

smart devices. In this paper, we propose a data collection

setup and machine learning models for joint human orienta-

tion and activity recognition using WiFi signals from a single

access point (AP) or multiple APs. The results show feasibil-

ity of joint orientation-activity recognition in an indoor envi-

ronment with a high accuracy.

Index Terms—Activity recognition, channel state infor-

mation, human-machine interaction, machine learning, WiFi.

1. INTRODUCTION

Human activity recognition (HAR) refers to detection and

recognition of human gestures and activities in an environ-

ment. Some major systems/mediums for collecting data are

wearable sensors (e.g. gyroscope and accelerometer), cam-

eras (e.g. still image and video), and radio frequency signals

(e.g. WiFi signals) [1]. HAR with wireless signals has at-

tracted attention due to its privacy preserving nature, broad

sensing coverage, and ability to sense the environment with-

out line-of-sight (LoS) [2]. This is particularly interesting

since the WiFi 802.11bf standard will enable remote moni-

toring and sensing [3].

Channel state information (CSI) in a wireless communi-

cation system can provide properties about the wireless chan-

nel and how a subcarrier has been affected in the environment.

Changes in the environment such as walking, falling, and sit-

ting can affect the CSI signals which can be used for various

sensing applications. CSI is measured in the baseband and is

a vector of complex values. A multiple-input multiple-output

(MIMO) wireless system provides a spatial diversity which

can be used for wider and more accurate sensing and detec-

tion of activities. This property of wireless signals can be

very useful in designing systems for human-machine inter-

action. Some examples are presence detection [4], security

systems [5], localization [6], and internet of things [7].

Various approaches have been proposed for HAR using

machine learning. That includes random forest (RF) [8],

hidden Markov model (HMM) [8], long-short-term mem-

ory (LSTM) [8], sparse auto-encoder (SAE) network [9],

attention-based bi-directional LSTM [10], and diversiﬁed

deep ensemble learning (WiARes) [11]. Most of the proposed

methods are based on training many trainable parameters for

feature extraction from CSI measurements. This approach

requires large CSI training data and hyper-parameter tuning.

In addition, most of these models due to their high compu-

tational complexity may not be suitable for implementation

on resource-limited devices such as smart phones and edge

devices [12]. LiteHAR [2] method uses a large number of

random convolution kernels without training them [13] for

feature extraction, followed by a pool of Ridge regression

classiﬁers per frequency for activity recognition. This ap-

proach enables fast and accurate HAR using CSI.

To the best of our knowledge, none of the previous works

in HAR have explored the possibility of predicting both ac-

tivity and orientation of the user using CSI. In this paper, ma-

chine learning models for prediction of the joint user activity

and orientation are introduced. Orientation prediction is par-

ticularly important for interaction with devices in smart en-

vironments, where multiple devices exist. It governs which

device the user is trying to interact with. We have built an

infrastructure for CSI measurements collection from multiple

access points (APs). Based on our previous work for a light-

weight HAR [2] solution, the idea of using 1-dimensional ran-

dom convolution kernels in [14] is utilized for feature extrac-

tion from CSI measurements. Then, Ridge regression classi-

ﬁers are used for prediction of the activation and orientation

of the user. The proposed models are evaluated for single AP

and multiple AP scenarios and the performance results are

discussed.

arXiv:2210.05078v2 [cs.HC] 21 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

JOINTHUMANORIENTATION-ACTIVITYRECOGNITIONUSINGWIFISIGNALSFORHUMAN-MACHINEINTERACTIONHojjatSalehinejad1;2,Member,IEEE,NavidHasanzadeh2,RadomirDjogo2,andShahrokhValaee2,Fellow,IEEE1KernCenterfortheScienceofHealthCareDelivery,MayoClinic,Rochester,MN,USA2DepartmentofElectrical&ComputerEngineering,Univer...

展开>> 收起<<

JOINT HUMAN ORIENTATION-ACTIVITY RECOGNITION USING WIFI SIGNALS FOR HUMAN-MACHINE INTERACTION Hojjat Salehinejad12 Member IEEE Navid Hasanzadeh2Radomir Djogo2 and.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

JOINT HUMAN ORIENTATION-ACTIVITY RECOGNITION USING WIFI SIGNALS FOR HUMAN-MACHINE INTERACTION Hojjat Salehinejad12 Member IEEE Navid Hasanzadeh2Radomir Djogo2 and

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: