Transcript Slide 1

Abstract
Algorithm
Algorithm
Fall-related injuries pose a serious threat to the health and
autonomy of older adults. As wearable fall detection systems are
often not adequate in reporting falls, we implement a 2D radio
tomographic imaging system for use in environmental fall
detection.
We locate people by implementing kernel-based radio
tomographic imaging [3]. This involves obtaining the received
signal strength values (RSS) for each transmitter/receiver pair.
These values are used to construct a short-term histogram, which
reflects recently acquired values, and a long-term histogram,
which acts as a baseline RSS distribution. The magnitude of the
kernel distance between these histograms determines the
attenuation on the link, which is used to reconstruct the
attenuation in each pixel. The location of a person is estimated as
the pixel with the maximum attenuation.
This system can be used in conjunction with a track-before-detect
approach to determine if a person has fallen. Alternatively, the
system can be extended to image 3D areas and detect falls with
a hidden Markov model [1]. In our implementation, we introduce a
simultaneous transmit and receive method that allows a very fast
frame rate, facilitating the implementation of a 2D fall detection
system.
Results
No object in line of sight
Object in line of sight
Baseband power spectrum of a three unit system
We effectively detected a person’s location with our
implementation of kernel-based radio tomographic imaging and
achieved a frame rate of over 20/sec. We created a visualization
that weights each pixel’s color according to the value of its pixel
attenuation estimator [4]. The pixel whose estimator has the
highest attenuation corresponds to the location of the person.
Pixels in line of sight of two transceivers
Implementation
Mid-position
20kHz
25kHz
30kHz
35kHz
Test of 2D RTI system
Visual output of imaging algorithm
Further Work
Standing
Lying
Hidden Markov model states for 3D RTI
65kHz
40kHz
Introduction
Thirty percent of people over the age of 65 fall each year; often
they are alone in their homes [2]. In the event of a fall, 80% of
people over the age of 90 with wearable help button devices do
not push the button, frequently because they are not wearing the
device. Device-free fall detection systems, such as those based
on radio tomographic imaging (RTI), alleviate this concern.
Since the human body attenuates many of the radio frequency
(RF) waves passing through it, RTI uses measurements of the
relative power of received RF waves to locate a person. 3D RTIbased fall detection detects the state of the person’s position
(lying, sitting, or standing) and decides whether the transition
between states occurred quickly enough to likely be caused by a
fall.
There are many possible extensions of this 2D RTI system. A
track-before-detect system based on particle filtering will enable
tracking and basic fall detection. A second layer of USRPs at ankle
height can be added for 3D RTI and a more accurate fall detection
using a 3-state hidden Markov model. There is room to investigate
extending the sensitivity region and accuracy of the links [5] and
controlling for multipath interference [6].
References and Acknowledgments
60kHz
55kHz
50kHz
45kHz
Visual output of imaging algorithm
We use 10 NI USRP-2920 units with full duplex daughterboards
to simultaneously transmit and receive. We chose a carrier
frequency of 1.4GHz, which attenuates well in a person’s body.
Each unit transmits at a different baseband frequency with 5kHz
spacing in between each signal. This prevents overlap between
the transmitted frequencies, as the USRPs do not have
synchronized internal oscillators.
[1] B. Mager, N. Patwari and M. Bocca, “Fall Detection Using RF Sensor Networks”, IEEE PIMRC, 2013
[2] C. Todd and D. Skelton, “What are the main risk factors for falls among older people and what are
the most effective interventions to prevent these falls?” WHO Regional Office for Europe, Tech. Rep.,
2004.
[3] Y. Zhao, N. Patwari, J. M. Phillips, and S. Venkatasubramanian. Radio tomographic imaging and
tracking of stationary and moving people via kernel distance. ACM ISPN, 2013
[4] Patwari, Neal, and Piyush Agrawal. "Effects of correlated shadowing: Connectivity, localization, and
RF tomography." ACM IPSN, 2008. International Conference on 22 Apr. 2008: 82-93.
[5] Kaltiokallio, Ossi, Hüseyin Yiğitler, and Riku Jäntti. "A Three-State Received Signal Strength Model
for Device-free Localization." arXiv:1402.7019 (2014).
[6] Wang, Zhenghuan et al. "Multichannel RSS-based Device-Free Localization with Wireless Sensor
Network." arXiv:1403.1170 (2014).
Thank you Jichuan Li, Ed Richter, Dr. Robert Morley, and Dr. Arye Nehorai
for all the indispensable assistance and guidance provided.