Lateral Fall Detection via Events in Linear Prediction Residual of Acceleration

Lateral fall is a major cause of hip fractures in elderly people. An automatic fall detection algorithm can reduce the time to get medical help. In this paper, we propose a fall detection algorithm that detects lateral falls by identifying the events in the Linear Prediction (LP) residual of the acceleration experienced by the the body during a fall. The acceleration is measured by a triaxial accelerometer. The accelerometer is attached to an elastic band and is worn around the test subject’s waist. The LP residual is filtered using a Savitzky-Golay filter and the maximum peaks are identified as falls. The results indicate that the lateral falls can be detected using our algorithm with a sensitivity of 84% when falling from standing and 90% when falling from walking.