Development of a Fall Detection System Based on a Tri-Axial Accelerometer

In this paper, we present a threshold-based fall detection system that uses data from a tri-axial accelerometer to differentiate falls from activities of daily life (ADL). Many previous studies on the subject use the sum vector (SV) of acceleration to identify the fall events. However, the SV curves of some daily activities (like jumping and running) are found to be like those of a fall. Therefore, a second parameter, which is attitude variation, is considered here to better distinguish a fall from an ADL. To evaluate the system's performance experimental data is used. The data is composed of acceleration signals collected from 21 falls and 21 ADL. Based on this data, the system's sensitivity is found to be around 95.23% and its specificity around 100%, which makes the proposed system successful in distinguishing falls from ADL. The results found in this study reveal that the proposed system's performance is comparable to previous studies found in the literature.