Hassle Free Fitness Monitoring

1
Hassle Free Fitness Monitoring

• David Jea, Jason Liu,
• Thomas Schmid, Mani Srivastava

2
Pervasive Health Care Systems

• Fitness Monitoring is the most Fundamental
  Functionality of Pervasive Health Care Systems
• Provides 24X7 Fitness Monitoring
• Sensor devices are clipped on the body
• Proactively Record changes in vital signs such as
  weight and blood pressure
• Appropriate Medical Services provided on the
  basis of recorded data

3
Challenges

• Privacy
• Security
• Finding a perfect balance between usability,
  privacy and security

4
Problems

• Large number of Devices hooked on the body
• Multiple type of sensors
• Privacy concerns at workplaces

5
Security Issues

• Network Security Issues
• User authentication issues
• Security problems related to stolen palmtops or
  PDAs

6
The Idea

• Build Fitness monitoring system for healthy
  individuals in a workplace
• Identification of the individual by only
  utilizing imprecise biometrics and existing
  information
• Maintaining the devices original user interface
• No additional sensors incorporated in the system

7
Design Guidelines

• Privacy
• Recorded data cannot be used as hard evidence
  (in court) to pinpoint exactly who the user is
• Feasibility
• The system is allowed to use existing information
• Usability
• Restoring the original interface of the device so
  that people of all age groups know how to use it

8
The Design
Possible Candidates
Biometric Matcher
Imprecise Physiological Info
Uncertainty Reduction
Context Reasoning
User Identity
Activity Information
9
Implementation

• The system consists of a weight scale and a blood
  pressure monitor
• Both devices communicate with the laptop
• Software program installed on laptop continuously
  record data and attach a timestamp to weight and
  blood pressure readings
• Facility for a user to input his/her name is also
  provided
• This step is to establish ground truth for the
  experiment

10
Inference Engine Components

• Biometric Matcher
• It implements a Bayes classifier that combines
  multiple sensor observations
• It assumes that each observation is unique
• This results in the identity of the subject
• Context Reasoning
• It is based on Reified Temporal Logic
• It provides with the users context
• It uses two meta-Predicates to express when
  things are true

11
Analysis
User Similarity in Physiological Information Seat in Lab Usage Habit Usage Habit Usage Habit
User Similarity in Physiological Information Seat in Lab Weight Scale BP Monitor Both
A Light V V
B They have similar weights. The differences in mean are less than 1.9 lbs V V V
C They have similar weights. The differences in mean are less than 1.9 lbs V V
D They have similar weights. The differences in mean are less than 1.9 lbs V V
E Their Difference in average weights is 1.1 lbs V V
F Their Difference in average weights is 1.1 lbs V
G Their Difference in average weights is 1.1 lbs V V
H Their Difference in average weights is 1.1 lbs V V
I Heavy V
12
Results for one Physiological Information
Physiological Data for Classifier Positive Match False Match
Weights 57.23 45.77
Systolic Blood Pressure 22.02 77.98
Diastolic Blood Pressure 43.90 56.10
Heartbeat Rate 25 75
13
Results based on Multiple Sources
Biometric matcher that combines all 4 physiological sources. Positive Match False Match
Classification Results for partial or complete data points. 77.9 22.1
Classification Results for complete data points only. 87.3 12.7
14
The accuracy of the context reasoning component
Context Reasoning Component Positive False Positive
The presence of a user based on network activity 89.47 10.53
15
Combining the biometric matcher with the context
reasoning.
Biometric Matcher only Biometric Matcher and Context Reasoning
Accuracy 78.16 83.80
16
Conclusion

• Built a health monitoring system which is hassle
  free
• Less privacy concerns
• No extra sensors hooked on the body
• Easy to Use
• Widely used by population
• How to handle uncertain usage?