IEEE Computational Intelligence Magazine - November 2019 - 32

I. Introduction

owadays, the ever-changing informatics technology
urges people living in modern society to be in a
tightly bounded state with a computer. While this
coherent state facilitates the accelerating rhythm of
urban life and work efficiency, it has gradually tied people to
chairs and deprived their exercise time. Its side effects like
Sedentary Behavior (SB) pose great threats to people's wellness [1]. Therefore, understanding user behavior, like knowing
whether the user is working, gaming or surfing and how long
s/he has been doing it, emerges as a key to spot such health
risk factors. Moreover, it constitutes a promising enabler to
many other fields like Human-Computer Interaction (HCI)
and Affective Computing (AC).
In general, there are two types of approaches to understanding user behavior. The traditional vision-based approach
leverages video as the main source, where cameras are
deployed to record and infer user behavior [2]-[4]. The visionbased approach is effective because of the mature Computer
Vision (CV) technology. However, it makes users to concern
about their privacy. Also, inherent defects of CV like line-ofsight and illumination constraints further jeopardize its usage
in practice. Sensor is another typical source, where wearable
sensors are attached to the human body to capture body gestures and deduce the corresponding user behavior [5]-[7].
Sensor has a limited sensing range, and thus multiple sensors
are needed at different parts of the human body to ensure
complete coverage of user gestures. Unfortunately, such
deployment is obtrusive for the user, hindering its practicability in real-world scenarios.
In this article, we introduce the WiFi signal, which is insensible to users, as an alternative source to vision and sensor for
perceiving user behavior. The key reason behind is that the
human body reflects or absorbs the WiFi signal, and thus
changes the WiFi Channel State Information (CSI) [8]-[10].
The inherent research problem is how to exploit WiFi CSI
that contains rich behavior information to retrieve micro-gestures like keystrokes and mouse movements for understanding
the corresponding user behavior?
Our response to the question is three-fold. Firstly, we explore signal processing to improve the sensing granularity of
WiFi CSI. In particular, we build a Fresnel-zone based model
to guide the antenna deployment to enhance minor signal
changes caused by user's micro-gestures. Then, we design a
light-weight segmentation algorithm to extract the micro-gestures from WiFi CSI automatically. The basic idea is to isolate
the signal fluctuations caused by the micro-gestures using the
stationary state as the benchmark. Lastly, we leverage computational intelligence to recognize these micro-gestures and then
understand their corresponding user behavior. We prototype
BeSense with low-cost WiFi devices and verify its performance in real environments. Extensive experiments demonstrate
that BeSense is very effective in capturing and recognizing
user's micro-gestures as well as understanding the corresponding user behavior.

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2019

We summarize the main contributions of this article as follows.
❏ We introduce the WiFi signal as a new source for user

behavior analysis. To the best of our knowledge, this is the
first work to recognize user behavior using WiFi.
❏ We leverage signal processing to build a Fresnel-zone based
model to enhance insignificant signal changes caused by
micro-gestures of users. Then we design a light-weight segmentation algorithm to extract micro gestures from the
continuous signal. Lastly, we exploit computational intelligence to recognize these micro-gestures as well as their corresponding user behaviors.
❏ We prototype BeSense on low-cost WiFi devices and verify
its performance in real environments. We also study alternative experimental materials for the human body to accommodate broader real-world scenarios.
The remainder of this paper is organized as follows: in the
next section, we provide an overview of the related works. We
introduce the system design in section III, and evaluate the
experimental results in section IV. We discuss the effect of
different objects on the received CSI signal in section V.
Finally, we conclude our work and discuss some open issues
in section VI.
II. Related Works

WiFi-based behavior sensing technology has many advantages
over traditional behavior sensing technology (e.g., visionbased sensing technology, infrared-based sensing technology
and dedicated sensor based sensing technology) in terms of
non-line-of-sight, passive sensing (no need to carry sensors),
low cost, easy deployment, no restrictions on lighting conditions, and strong scalability. A large number of applications
have emerged based on existing WiFi signals. From daily
behavioral awareness [11], [12] and gesture recognition [13],
[14] to identity authentication [15], [16] and from individual
physiological indicators [17], [18] to group perception [19],
[20] and fall detection [21], [22], behavior sensing technology
based on WiFi is showing unprecedented potential for application, achieving not only the interaction between machines
and machines but also the natural interaction between
humans and machines.
In the early days, WiFi-based behavior sensing mainly uses
Received Signal Strength (RSS). Sigg et al. [23] use a software
radio to transmit RF signals and determine human motion
based on changes in RSS. Abdelnasser et al. leverage RSS to
identify 7 different gestures [13] and respiratory detection [24].
We also built a similar RSS-based system PAWS to handle
whole-body activities [8]. Since RSS is coarse-grained and CSI
can yield more detailed information, recent research mainly
uses CSI for behavior sensing. WiFall [25] uses CSI to implement the fall detection system. Zeng et al. [19] leverage CSI to
recognize five different customer behavior states. We also built a
CSI-based system MoSense to pinpoint the motions in a realtime manner [9].
These schemes basically rely on high-level features like
location, velocity and direction of a motion for behavior

IEEE Computational Intelligence Magazine - November 2019

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2019