Transcript Slide 1

UC BERKELEY
UC DAVIS
UC MERCED
UC SANTA CRUZ
Eldercare April 2006
Ruzena Bajcsy, Mike Eklund, Shankar Sastry, Steve Glaser
Presentation by Ruzena Bajcsy
1
White House Forum: Technologies
for Successful Aging (10/4-5/00)
 Number of senior in U.S. estimated to increase
from over 33 million today to:
» 53 million in 2020
» 77 million in 2040
 By 2023, the demographic profile for the nation will
be similar to the profile in Florida today.
2
Parent Support Ratio:
1950-2050
As the number of elderly needing care increases, the number of
potential caregivers decreases.
Today, 1 in 4 U.S.
families care for an
older adult.
30
25
20
15
No. +85
10
Persons 85+ per
100 people
50-64 years old
5
0
1950
1990
2010
2030
By 2005, nearly
40% of U.S. workers
will be more
concerned caring for
a parent than a child.
2050
Source: U. S. Census
3
Squaring the U.S. Population Pyramid
1950-2030
Age
85+
80-84
75-79
70-74
65-69
60-64
55-59
50-54
45-49
40-44
35-39
30-34
25-29
20-24
15-19
10-14
5-9
0-4
1950
1980
2000
2030
(150,216,000)
(227,658,000)
(267,955,000)
(304,807,000)
4
Opportunities and Challenges
 Technologies that meet the challenges of aging will
be increasingly valuable
 We must identify collaborative, technology transfer,
technology development and deployment
opportunities for government, industry, and
academia that help improve the independence,
mobility, security, and health of aging U.S. citizens
 We must examine potential opportunities and
barriers and identify and prioritize
recommendations both near and long term,
including Grand Challenge class
recommendations
5
Focus on Fall detection
 While our agenda is much broader , that is we are
interested in a comprehensive program of
integrating sensory, wireless technology ,
embedded systems into an architecture that would
facilitate monitoring the elderly population
 Their medical records s well as their daily
activities.
 In this presentation we shall discuss only one
aspect that is the Fall detection and alarm system.
6
Trend #2
Worldwide healthcare crisis is here
- Every major world economy has health as biggest percentage
- Nursing shortage in many parts of the world
- South Korea and Japan technology infrastructure
7
Trend #3
Elder care is returning home again
Poor Houses / Almshouses
“pauper”
Home
“grandma”
Only way to save costs but
increase quality is home care.
Insane Asylum
“inmate”
Home care is fastest growing
segment of health industry.
Home
“grandma”
Assisted Living
“resident”
Hospital
“patient”
Nursing Home
“senior citizen”
8
Trend #4
Convergence is actually happening
• Everyday health through everyday devices
• Everything is a touchpoint to everything else
• Every device has a chip, every chip as a radio
9
1. Proactive Health Lab in various
Institutions
http://www.intel.com/research/prohealth/
10
SensorNet Overview
Sensors
E.g. Bluetooth
Sender
Fall Detector
Mobile Gateway
Hospital
Terminal, WLAN
Internet
and/or
telephone
Mobile Phone
Integrated
Camera
RS-232
E.g. Bluetooth
Sender
Berkeley
Mote
RS-232
Berkeley
Motes
Berkeley
Mote
Sensors
Home Health
System
Zigbee
11
Security and Privacy: Wireless
 Bluetooth has built in (and evolving) security.
» There are three modes of security for Bluetooth access
between two devices.
• Security Mode 1: non-secure
• Security Mode 2: service level enforced security
• Security Mode 3: link level enforced security
 ZigBee (802.15.4) security includes methods for
»
»
»
»
key establishment,
key transport,
frame protection, and
device management.
12
Wireless Network: Capabilities
 E.g., Wireless video consultation
» Uses local Bluetooth, 802.11 network and/or ethernet
» Monitor sensors, medical devices, etc.
Health Care
Professional
User, or Health Care
Professional, etc
FallSensors
Detector
Server
13
Fall Detector

Features:
»
»
»
»
»
»
»

3-axis, ±10 g accelerometers
on board GPS
Battery powered
RS-232 connection
C programmable
80 Hz sampling
4 hours recording,
or continuous streaming
Functionality:
» Record or stream accelerometer data for testing
» Embedded fall detection algorithms
» Connect to Bluetooth radio
• Bluegiga device
• Initiate a connection with a control device, i.e. the home or
mobile gateway
14
Basic Signal Processing
 Three steps:
» Calibration
» Filtering
• Remove transients
and noise
» Coordinate
Transformation
• For pose recovery and
orientation
Raw XYZ
Calibration
Low Pass 1
Low Pass 2
Coordinate
Transformation
Align to
World XYZ
Pose
Recovery
Motion
Classification
15
Pose Recovery (of device):
Filtering & Transformation
 Gravity is always present
» Use low pass filters to reveal
the Earth frame of reference
 Transform to Spherical
Coordinates to reveal gravity
and orientation
Angle off
vertical
(rad) Φ
R  x2  y 2  z 2
  arctan  y x 
  arctan

x2  y 2 z
XYZ Data, filtered and not
Gravity

Rotation θ
Spherical Data, filtered and not
16
The Experiment in Sonoma
 We have conducted 3 experiments each with 3-5
people wearing our device for approximately two
hours.
 First we have instructed the persons to perform
repeated sit-down stand-up, walk on stairs, walk
on floor, each for 5 times during which we
recorded the acceleration.
 Second we left the person to do normal activities
while wearing the sensor for approximately 1 ½
hours.
17
Activities of Interests
 Activities of Daily Living
»
»
»
»
Walking
Sitting
Standing
Other normal activities
 Falls
» Categorizing the types and severities
18
Example Results
 The recorded data from
accelerometer based fall
sensor is analyzed and
replayed with this Matlab
program.
 Examples
» an elderly woman walking
with a walker (top right),
» a woman sitting down (top
left),
» a woman standing up
(bottom right)
» a man stretching his arm
(bottom left)
19
Example of Identifying ADLs: Sitting
Heavy lines represent average results in X, Y, Z
20
Example of Identifying ADLs: Standing
Heavy lines represent average results in X, Y, Z
21
Example of Identifying ADLs: Walking
Heavy lines represent average results in X, Y, Z
22
Comparing: falling and sitting
X Accel
X Accel
Y Accel
Y Accel
6
2
6
2
4
1
4
1
2
0
2
0
0
-1
0
-1
-2
0
100
200
300
-2
-2
0
100
Z Accel
200
300
400
600
800
-2
0
200
400
600
800
600
800
Norm
3
8
8
2
2
6
6
1
1
0
0
4
4
-1
-1
2
2
-2
-2
-3
200
Z Accel
Norm
3
0
0
100
200
300
0
-3
0
100
200
Falling (Trained Judo-ist)
300
0
200
400
600
800
0
0
200
400
Sitting (Septuagenarian)
23
Lessons learned
 While the signal for sit-down and stand-up a clear
distinguishable characteristics, it is also clear that
there are substantial differences from person to
person that need to be accounted in order to avoid
false positives.
 This suggest that the data must be normalized with
respect to the weight and height of the person.
24
But what about in-between?
 False positives
» Unnecessary alerts, worries
» At the least very bothersome
 False negatives
» Critical problem
 Solution must be very robust:
»
»
»
»
Good algorithms and devices
User interaction
Sensor fusion
Reliability
25
Respect for Privacy
 As we interviewed the participants we learned that
if they indeed would have fallen, they wish to alert
» first the neighbors
» then the children/relatives
» and only later the 911
26
Q&A
27