Wireless Insole For Activity Recognition

Foot stress can cause ulceration for diabetic individuals, and ulcers, in turn, can cause reduced quality of life and increased risk for lower extremity amputation. The goal of this project is to develop an unobtrusive device that acts as an early warning system to prevent excessive foot stress and its debilitating effects. Existing in-shoe measurement devices are either too expensive, restricted to the laboratory, or else insufficient with regard to data storage. Novel analysis techniques are also needed to handle vast amounts of data collected during a multitude of activities.

The photo below shows the inside of our prototype insole, with several force-sensing resistors that enable measurement of plantar foot pressure in key locations. The inset shows our portable microcontroller, which can either log data to a CompactFlash card or else transmit measurements to a host computer via Bluetooth.

insole w/ microcontroller

For his senior honors project at UMass, Subie Patel (now at Stanford) used the device to recognize various activities such as walking on stairs and around corners. Approximately 20 features were used to describe the relative timing, amplitude, and width of various peaks in the segmented pressure time series. A decision tree algorithm was then used to learn a classifier for seven prespecified activities. Although more sophisticated methods are needed to develop a comprehensive system, the preliminary results are promising as illustrated by the confusion matrix below.

Correctly classified instances    266   91.4 %
Incorrectly classified instances   25    8.6 %

=== Confusion Matrix ===

  a  b  c  d  e  f  g   <-- classified as
 81  3  0  0  0  1  3 |  a = walking
  1 39  0  0  0  0  0 |  b = walkingBackwards
  0  0 39  5  0  0  0 |  c = climbingUp
  0  0  1 54  0  0  0 |  d = climbingDown
  0  0  0  0 29  0  1 |  e = still
  3  0  0  0  0 12  2 |  f = turnLeft
  4  0  0  0  0  1 12 |  g = turnRight

updated 12-Dec-2005