Understanding and Modeling of WiFi Signal Based Human Activity Recognition Wei Wangt Alex X.Liutt Muhammad Shahzad:Kang Ling' Sanglu Lut tState Key Laboratory for Novel Software Technology,Nanjing University,China +Dept.of Computer Science and Engineering,Michigan State University,USA ww@nju.edu.cn,(alexliu,shahzadm)@cse.msu.edu,lingkang@smail.nju.edu.cn,sanglu@nju.edu.cn Abstract limited operation range of just tens of centimeters [2].Wearable Some pioneer WiFi signal based human activity recognition sys- sensors based approaches are inconvenient sometimes because of tems have been proposed.Their key limitation lies in the lack of the sensors that users have to wear.Recently.WiFi signal based hu- a model that can quantitatively correlate CSI dynamics and human man activity recognition systems,such as WiSee [17].E-eyes [27]. activities.In this paper.we propose CARM,a CSI based human and WiHear[26],have been proposed based on the observation that Activity Recognition and Monitoring system.CARM has two the- different human activities introduce different multi-path distortions oretical underpinnings:a CSI-speed model,which quantifies the in WiFi signals.WiSee uses USRP to capture the OFDM signals correlation between CSI value dynamics and human movement and measures the Doppler shift in signals reflected by human bod- speeds,and a CSI-activity model,which quantifies the correlation ies to recognize nine gestures.E-eyes uses Channel State Inform- between the movement speeds of different human body parts and ation(CSI)histograms as fingerprints for recognizing daily human a specific human activity.By these two models,we quantitatively activities such as brushing teeth.WiHear uses specialized direc- build the correlation between CSI value dynamics and a specific tional antennas to obtains CSI variations caused by lip movement human activity.CARM uses this correlation as the profiling mech- for recognizing spoken words.Their key advantages over camera anism and recognizes a given activity by matching it to the best-fit and sensor based approaches are that they do not require lighting, profile.We implemented CARM using commercial WiFi devices provide better coverage as they can operate through walls,preserve and evaluated it in several different environments.Our results show user privacy,and do not require users to carry any devices as they that CARM achieves an average accuracy of greater than 96%. rely on the WiFi signals reflected by humans Categories and Subject Descriptors 1.2 Limitations of Prior Art The key limitation of these pioneer WiFi based human activity C2.1 [Network Architecture and Design]:Wireless communica- recognition systems is the lack of a model that can quantitatively tion correlate CSI dynamics and human activities.As such,these sys- General Terms tems mostly rely on the statistical characteristics of WiFi signals such as Doppler movement directions and distributions of signal Experimentation,Measurement strength,to distinguish different human activities.The lack of such a model limits the further development of WiFi based human activ- Keywords ity recognition technologies.Without such a model,it is difficult Channel State Information(CSD);WiFi;Activity Recognition; to understand the correlation between WiFi signal dynamics and human activities.Furthermore,without such a model,it is diffi- 1.INTRODUCTION cult to optimize the performance of such systems due to the lack of adjustable parameters,and we have to resort to trial-and-error for 1.1 Motivation performance optimization. Human activity recognition is the core technology that enables a 1.3 Proposed Approach wide variety of applications such as health care,smart homes,fit- ness tracking.and building surveillance.Traditional approaches In this paper,we propose CARM,a CSI based human Activity use cameras [6],radars [2],or wearable sensors [7,33].How- Recognition and Monitoring system.CARM consists of two Com- ever,camera based approaches have the fundamental limitations mercial Off-The-Shelf(COTS)WiFi devices as shown in Figure 1,one for continuously sending signals,which can be a router. of requiring line of sight with enough lighting and breaching hu- and one for continuously receiving signals,which can be a laptop. man privacy potentially.Low cost 60 GHz radar solutions have When a human activity is performed in the range of these two devices,on the WiFi signal receiver end,CARM recognizes the Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed human activity based on how the CSI value changes.CARM has for profit or commercial advantage and that copies bear this notice and the full cita- two theoretical underpinnings that we propose in this paper:a tion on the first page.Copyrights for components of this work owned by others than CSI-speed model and a CSl-activity model.Our CSI-speed model ACM must be honored.Abstracting with credit is permitted.To copy otherwise,or re- quantifies the correlation between CSI value dynamics and human publish,to post on servers or to redistribute to lists,requires prior specific permission movement speeds.Our CSI-activity model quantifies the correla- and/or a fee.Request permissions from Permissions@acm.org. MobiCom'/5,September 7-11,2015,Paris.France. tion between the movement speeds of different human body parts ©2015ACM.ISBN978-1-4503-3619-2/1509.S15.00 and a specific human activity.By these two models,we quantitat- D0 http:ldx.doi.org/10.11452789168.2790093. ively build the correlation between CSI value dynamics and a spe-Understanding and Modeling of WiFi Signal Based Human Activity Recognition Wei Wang† Alex X. Liu†‡ Muhammad Shahzad‡ Kang Ling† Sanglu Lu† †State Key Laboratory for Novel Software Technology, Nanjing University, China ‡Dept. of Computer Science and Engineering, Michigan State University, USA ww@nju.edu.cn, {alexliu,shahzadm}@cse.msu.edu, lingkang@smail.nju.edu.cn, sanglu@nju.edu.cn Abstract Some pioneer WiFi signal based human activity recognition sys￾tems have been proposed. Their key limitation lies in the lack of a model that can quantitatively correlate CSI dynamics and human activities. In this paper, we propose CARM, a CSI based human Activity Recognition and Monitoring system. CARM has two the￾oretical underpinnings: a CSI-speed model, which quantifies the correlation between CSI value dynamics and human movement speeds, and a CSI-activity model, which quantifies the correlation between the movement speeds of different human body parts and a specific human activity. By these two models, we quantitatively build the correlation between CSI value dynamics and a specific human activity. CARM uses this correlation as the profiling mech￾anism and recognizes a given activity by matching it to the best-fit profile. We implemented CARM using commercial WiFi devices and evaluated it in several different environments. Our results show that CARM achieves an average accuracy of greater than 96%. Categories and Subject Descriptors C2.1 [Network Architecture and Design]: Wireless communica￾tion General Terms Experimentation,Measurement Keywords Channel State Information (CSI);WiFi; Activity Recognition; 1. INTRODUCTION 1.1 Motivation Human activity recognition is the core technology that enables a wide variety of applications such as health care, smart homes, fit￾ness tracking, and building surveillance. Traditional approaches use cameras [6], radars [2], or wearable sensors [7, 33]. How￾ever, camera based approaches have the fundamental limitations of requiring line of sight with enough lighting and breaching hu￾man privacy potentially. Low cost 60 GHz radar solutions have Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full cita￾tion on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re￾publish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from Permissions@acm.org. MobiCom’15, September 7–11, 2015, Paris, France. c 2015 ACM. ISBN 978-1-4503-3619-2/15/09 ...$15.00. DOI: http://dx.doi.org/10.1145/2789168.2790093. limited operation range of just tens of centimeters [2]. Wearable sensors based approaches are inconvenient sometimes because of the sensors that users have to wear. Recently, WiFi signal based hu￾man activity recognition systems, such as WiSee [17], E-eyes [27], and WiHear [26], have been proposed based on the observation that different human activities introduce different multi-path distortions in WiFi signals. WiSee uses USRP to capture the OFDM signals and measures the Doppler shift in signals reflected by human bod￾ies to recognize nine gestures. E-eyes uses Channel State Inform￾ation (CSI) histograms as fingerprints for recognizing daily human activities such as brushing teeth. WiHear uses specialized direc￾tional antennas to obtains CSI variations caused by lip movement for recognizing spoken words. Their key advantages over camera and sensor based approaches are that they do not require lighting, provide better coverage as they can operate through walls, preserve user privacy, and do not require users to carry any devices as they rely on the WiFi signals reflected by humans. 1.2 Limitations of Prior Art The key limitation of these pioneer WiFi based human activity recognition systems is the lack of a model that can quantitatively correlate CSI dynamics and human activities. As such, these sys￾tems mostly rely on the statistical characteristics of WiFi signals, such as Doppler movement directions and distributions of signal strength, to distinguish different human activities. The lack of such a model limits the further development of WiFi based human activ￾ity recognition technologies. Without such a model, it is difficult to understand the correlation between WiFi signal dynamics and human activities. Furthermore, without such a model, it is diffi- cult to optimize the performance of such systems due to the lack of adjustable parameters, and we have to resort to trial-and-error for performance optimization. 1.3 Proposed Approach In this paper, we propose CARM, a CSI based human Activity Recognition and Monitoring system. CARM consists of two Com￾mercial Off-The-Shelf (COTS) WiFi devices as shown in Figure 1, one for continuously sending signals, which can be a router, and one for continuously receiving signals, which can be a laptop. When a human activity is performed in the range of these two devices, on the WiFi signal receiver end, CARM recognizes the human activity based on how the CSI value changes. CARM has two theoretical underpinnings that we propose in this paper: a CSI-speed model and a CSI-activity model. Our CSI-speed model quantifies the correlation between CSI value dynamics and human movement speeds. Our CSI-activity model quantifies the correla￾tion between the movement speeds of different human body parts and a specific human activity. By these two models, we quantitat￾ively build the correlation between CSI value dynamics and a spe-