Pre-impact fall detection

Falls and fall-related injuries are among the most serious and common medical problems experienced by the elderly. Support could be given to prevent severe injuries from those falls. Therefore the design of a robotic assistive device is currently investigated in the MARS3-project. Such a device requires knowledge of the human posture to determine loss of balance. At the moment several methods are available to detect a fall, these are not accurate enough to be implemented into a robotic device. In this study a reliable trigger for a robotic device is investigated. During an experiment in which participants performed a balance task, human movement is recorded. Motion Capture is performed to track the posture changes and a sensor system is used to obtain input for the assistive device. It is analyzed that the output of the sensory system is in correspondence with the movements recorded by Motion Capture. Based on the Motion Capture the data is divided into stable and unstable phases. Six methods to design a classifier that is able to distinguish between stable and loss-of-balance are proposed. These methods are all trained and evaluated with the data sets obtained in the experiment. The main measures to evaluate the performance of the algorithms are sensitivity, i.e. detect all instances of loss of balance, and specificity, i.e. no false alarms. Although the sensory information is in correspondence with the actual body movement, the proposed classifiers do all result in sensitivity and specificity rates that are not high enough to implement into an assistive device. The value of information is too low, it is not possible to distinguish stable and unstable phases by means of the proposed classifiers.

Subject

Fall detection

To reference this document use:

http://resolver.tudelft.nl/uuid:3820aabf-2d9d-4c41-a29e-584b96435977

Embargo date

2019-09-22

Part of collection

Student theses

Document type

master thesis

Rights

Files

PDF

Thesis_Myrthe_Koning.pdf

3.11 MB

Close viewer