Home Health Care and Assisted Living

1
Learning Micro-Behaviors In Support of Cognitive
Assistance
John A. Stankovic, Leo Selavo, Anthony D. Wood
Department of Computer Science, University of
Virginia
AlarmNet is a wireless sensor network (WSN)
system for smart health-care that opens up new
opportunities for continuous health monitoring in
assisted-living or residential facilities. It
provides real-time (24/7) access to physiological
and environmental data, and tracks long-term
changes in behavioral patterns for cognitive
assistance by exploiting large numbers of cheap
sensors, cell phones, a high degree of
heterogeneity, and special purpose hardware.
I. System Overview
IV. Assistive Feedback

• Improving Quality of Life and Health Care
• Continuous, real-time monitoring
• Nutrition and hygiene
• Disease progression management
• Treatment compliance
• Environmental conditions
• Unobtrusive smart clothes
• Support longitudinal studies
• Detect at-risk medical situations
• Trigger alerts to health care providers
• Cognitive Assistance
• Use many, cheap, wearable sensors
• Detect individuals macro- micro-behaviors
• Use in-situ devices for interaction
• Maintain resident privacy and security

• Intelligent PDAs and Cell phones provide
• Enhanced sensing and monitoring
• Wide-area communications
• Cognitive assistance interaction
• Mobile personal storage
• Unique ID for data association

Pills
Assisted Living Facility
Video cameras
Motion sensors
Backbone nodes
Motes (emplaced WSN)
Wearable devices provide low-power,
easily-accessible displays for system feedback of
reminders and prompts, risk alerts, and current
sensor data.
II. Heterogeneous Sensing Hardware
Large numbers of cheap sensors provide better
activity classificationbeyond typical ADLsand
reveal where cognitive assitance is most needed.
Experimental Testbed
Body networks help record micro-behaviors for
activities such as walking, eating and stillness
using five 2-axis accelerometers embedded in a
jacket.
V. Behavior Patterns
A medical application monitors the Circadian
Activity Rhythms (CAR) of a resident to extract
high-level activity patterns and detect
behavioral anomalies.
III. Wireless Sensor Network
Continuous Monitoring
User Reminders
AlarmNet also records precise micro-behaviors
that constitute higher-layer behaviors. A
significant deviation from ones own norm may
indicate cognitive decline. Tailored cognitive
assistance can be offered to the resident, and is
stored in the database with privacy protections.
VI. Privacy Challenges
Ambient Sensing
Transience of Data. Most data collected by the
system are transient physiological and activity
data. They require real-time processing. Resource
Constraints. Despite the large amount of
transient data to be collected, WSN devices have
very limited power and storage. This requires
careful design of data aggregation and processing
algorithms. Cross-Boundary Addressing. Data is
streamed to back-end servers, where privacy
schemes control how data is stored and who can
access it. Policies must be consistent in the
back and front-ends of the system. Transience of
Privacy Preferences. Individuals privacy
preferences are dynamic, often depending on the
current context and health conditions. They
should be dynamic, adjustable, and adaptive in
emergencies.
Interaction and Viewing
Analysis and Feedback