Falls commonly occur in older adults and could result in long-lies when no one is around to assist, which could result to additional emotional and physical consequences. The use of inertial sensors allows a portable and unobtrusive way to detect motion, enabling the automatic detection of falls when used with a fall detection algorithm. The wrist and trunk are two locations that are favorable for fall detection as the former provides a convenient location for the user, while the latter provides a good location for capturing the body’s general motion. The objective of this thesis is to further improve the performance of a wrist-mounted and a trunk-mounted threshold-based fall detection algorithm using inertial sensors comprised of tri-axial accelerometer, tri-axial gyroscope, and a barometric pressure sensor. The algorithms were tested using a comprehensive set of laboratory-simulated falls, activities of daily living (ADL), and near-falls.

In the first study, a wrist-based fall detection algorithm for a commercially available smartwatch was proposed. The algorithm used forearm angle to filter the forearm’s downward vertical orientation that could be associated to a non-fall event’s post-activity position. Additionally, to deal with disturbance in barometric pressure data during dynamic motion, barometric pressure was used selectively in a Kalman filter. The algorithm gave 100% sensitivity, 97.2% ADL specificity, and 97.1% non-fall (i.e. including both ADLs and near-falls) specificity.

In the second study, the addition of either difference in altitude or average vertical velocity to a trunk-based algorithm that uses vertical velocity + vertical acceleration + trunk-angle (base algorithm) was investigated. The experimental results show that adding either difference in altitude or average vertical velocity was able to increase the algorithm’s non-fall specificity from 91.8% to 98.0% and 99.6%, respectively.