Investigation of the effects of body type on accelerometer based fall detection

Abstract

Precision or personalized medicine follows the idea that analyzing a patient's genome can help stratify patients into treatment groups. Patients with cancer, for example, are split into groups based on the genetic make-up, rather than the location, of their tumor and given medicine that corresponds to their specified treatment group. Instead of prescribing the same medicine for all patients with lung cancer, doctors are able to personalize treatment based on a person's genetics. Could fall detection algorithms benefit from this type of personalized approach? This project analyzes accelerometer data containing falls and activities of daily living gathered from 16 stunt actors and 20 elderly volunteers as well as 22 falls from the FARSEEING dataset with the goal of discovering whether algorithms personalized to a person's physiological features (such as height or weight) perform better than the traditional generalized approach. This thesis demonstrates that personalized algorithms can potentially make fall detection more efficient for some people. The unscripted test results show that the algorithms personalized to build and weight, decreased false alarms in several participants as compared to the overall algorithm. The FARSEEING data results show that algorithms, personalized to weight and sex, may increase the efficiency and accuracy of fall detection algorithms. An analysis of which physiological factors help determine whether a person needs a personalized or a general algorithm is needed to determine whether or not personalized algorithms make economic sense.