IEEE - Aerospace and Electronic Systems - November 2019 - 35

Shah and Fioranelli

Figure 6.
Sketch for three wandering behaviors (a) Typical scenario for random, pacing, and lapping gaits. (b) Experimental setup. (c) Time history of
three gaits considering amplitude information.

Wi-Fi-BASED KEY STROKE IDENTIFICATION
A keystroke recognition system leveraging channel state
information obtained using Wi-Fi signals was proposed by
Ali et al. [46]. The authors for the first time demonstrated
that WCI can be used to identify keystrokes of 37 available keys on keyboard such as 1 space bar, 10 digits, and
26 alphabets. The authors exploited the finger and hand
movement which carry unique signature in terms of formation and direction when the user is typing specific keys
that produce a unique wireless channel imprint as a result
generates specific CSI values known as CSI-waveforms.
This method uses the induced CSI values for each keystroke in their feature space. The state-of-the-art Wi-Fi
devices transmit data at higher data rates and are enough
to form a high-resolution CSI-data for specific keystroke.
The principal component analysis (PCA) was used to
eliminate the unwanted noisy components from the signals
received. In real-world application, this method can identify keystrokes continuously and can determine the sentences with an accuracy of 93%. However, the proposed
framework only works in a controlled setting and has
several limitations.

between 1 and 60 Hz. Finally, a weighted-moving-average
technique was used to further remove unwanted components form the CSI data. The CSI data received was
divided into small chunks to extract the desired features
for training and testing. The finger-grained gesture recognition leveraging Wi-Fi technology works well for single
person under stable environment (where no external
movement is observed) including laboratories where only
two occupant one subject and another taking reading is
present. This system achieved an average accuracy of
more than 90%. Transmitter and an omni-directional
antenna as receiver. The distance between the transceiver
pair was 7 m. As an example, the variances of amplitude
information for three wandering behaviors are presented
in Figure 7(c). Each gait induced a unique signature, i.e.,
for random behavior the amplitude information varied
between 3 and 25 dB, for pacing the amplitude information fluctuated around 25 dB and for lapping the data
change between 23 and 25 dB for subcarrier number 20. A
SVM was used to classify three different gaits with an
overall accuracy of 90%.

Wi-Fi-BASED HAR
FINGER-GRAINED GESTURE RECOGNITION
The finger-grained gesture recognition or Wi-Finger proposed in [47] obtained number and text as input using finger-grained gestures of nine digits using commercially
available low-cost Wi-Fi devices. The core idea was to
apply the number and text input for human-compute interaction. The proposed system involved three main stages.
Initially, a Hampel identifier was used to remove the outliers. Second, a low-pass filer such as Butterworth filter
was applied in order to eliminate frequencies ranging
NOVEMBER 2019

A noninvasive, low-cost indoor human activity recognition system was introduced by Wang et al. [48] known as
the E-eyes. The system exploited Wi-Fi signal using channel state information obtained for a group of 30 subcarriers in indoor settings [49], [50]. The system provided an
adequate solution where minimum wireless network infrastructure was required using commercially available, lowcost, off-the-shelf Wi-Fi devices. Contrary to the gesture
identification system, the HAR required loosely defined
set of motion over time, generally known as large-scale
body movements. As far as data processing for E-eyes is

IEEE A&E SYSTEMS MAGAZINE

IEEE - Aerospace and Electronic Systems - November 2019

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - November 2019