Discovering Activities of Daily Life Using RFID

1
Discovering Activities of Daily Life Using RFIDs

• Masters Thesis
• Sandor Dornbush

2
Overview

• What are activities of daily life and why do we
  care?
• How its been done in the past.
• How to use RFID to determine ADLs.
• How I will do it better.

3
Activities of Daily Life

• ADL
• Its what we do every day.
• Brush teeth
• Eat
• Sleep
• Etc.

4
Who Cares About ADLs?

• Mostly senior care.
• Gives children and care givers a way to remotely
  monitor a senior citizens life.
• Allows senior citizens to live independently
  longer.
• Also useful for smart spaces.

5
Overview

• What are activities of daily life and why do we
  care?
• How its been done in the past.
• How to use RFID to determine ADLs.
• How I will do it better.

6
Previous Techniques

• Machine Vision
• Imbed cameras into the living space.
• Try to extract meaningful information from the
  video stream.
• Very hard problem, low success rate.
• Motion Detectors
• Can only detect the region of the house that a
  person is in.
• Its possible to infer activities, but with low
  success rate.

7
Overview

• What are activities of daily life and why do we
  care?
• How its been done in the past.
• How to use RFID to determine ADLs.
• How I will do it better.

8
Philipose Et Al.

• Put RFID tags on important objects in the house.
• Create a wearable tag reader.
• Currently a glove. I imagine a bracelet being
  the final form.

• Can determine what objects a user has
  interacted with.

9
Determining ADLs from RFID Data

• How can one map from a stream of RFID tag
  readings to a concrete ADL?
• Philipose Et Al. Approach
• Parse natural language recipes into a machine
  representation of a series of objects.
• Combine those objects into a Bayesian Network.
• Derive the probabilities of the Bayesian Network
  from the Google API.

10
Overview

• What are activities of daily life and why do we
  care?
• How its been done in the past.
• How to use RFID to determine ADLs.
• How I will do it better.

11
Proposed Improvements

• Software Only
• I will not try to improve on the design of the
  glove reader.
• Learning Systems
• In the study the participants recorded their
  activities to check the accuracy of their system.
  These records can be used for supervised
  learning.
• Instead of Bayesian Networks from the Google API
  try a variety of techniques.
• Genetic Algorithms
• Learning Bayesian Networks
• Decision Trees
• Neural Networks

12
Personal Vs. General

• A learning system could learn the behaviors of
  the user.
• Personalization could provide higher accuracy.
• For example if a person drinks their tea black,
  interacting with the refrigerator should not
  indicate that they may be making tea.

13
Knowledge Representation

• Express the recipe for an ADL in a concise
  machine readable format.
• XML or RDF
• Provides more precise representation of the ADL.
• Allows for more attributes, such as duration of
  the ADL.