A two-threshold fall detection algorithm for reducing false alarms

Wireless health monitoring can be used in health care for the aged to support independent living, either at home or in sheltered housing, for as long as possible. The most important single monitoring need with respect to security and well-being of the elderly is fall detection. In this paper, a two-threshold MATLAB-algorithm for fall detection is described. The algorithm uses mainly tri-axial accelerometer and tri-axial gyroscope data measured from the waist to distinguish between fall, possible fall, and activity of daily living (ADL). The decision between fall and possible fall is done by the posture information from the waist- and ankle-worn devices ten seconds after the fall impact. By categorizing falls into these two sub-categories, an alarm is generated only in serious falls, thus leading to low false alarm rate. The impact itself is detected as the total sum vector magnitudes of both the acceleration and angular velocity exceeds their fixed thresholds. With this method, the sensitivity of the algorithm is 95.6% with the set of 68 recorded fall events. Specificity is 99.6% with the set of 231 measured ADL movements. It is further shown that the use of two thresholds gives better results than just one threshold.