A real-time sitting posture tracking system
Abstract
As computing becomes more ubiquitous, there is a need for distributed intelligent human-computer interfaces that can perceive and interpret a user's actions through sensors that see, hear and feel. A perceptually intelligent interface enables a more natural interaction between a user and a machine in the sense that the user can look at, talk to or touch an object instead of using a machine language. Although research on haptic (i.e., touch-based) interfaces has received less attention in the past as compared to that on visual and auditory interfaces, it is emerging as a new interdisciplinary field that holds much promise for the future. The goal of the sensing chair project is to enable a computer to track, in real time, the sitting postures of a user through the use of surface-mounted contact sensors. Given the similarity between a pressure distribution map from the contact sensors and a gray-level image, we have adapted computer vision and pattern recognition algorithms for the analysis of sitting pressure data. We collected a database containing samples of sitting pressure data for ten commonly used postures from 40 subjects. This database was used to train and test posture classification systems. The first system used an eigenspace-based approach for posture classification. It was found that the overall classification rate was 96 percent correct on test samples taken from subjects who had contributed training data. The accuracy rate dropped to approximately 79 percent when tested on data from subjects who did not contribute training samples. The original system was modified into a two-stage classification system, (using either a Bayesian classifier or one that uses a pyramid representation) having been tested. With the addition of the second classifier, the overall accuracy was increased by approximately five percent to 84 percent correct. Classification accuracy for individual posture classes was as high as 90.4 percent.
Degree
Ph.D.
Advisors
Tan, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.