Deep Learning Methods for Fall Detection

Remote and passive monitoring of elderly and patients with neurodivergent diseases has become popular and useful in modern medicine. This technology has the ability to provide these patients a degree of autonomy over their lives, diminishing the necessity for a constant caregiver in their home. This paper outlines and targets the fall detection topic using a triaxial accelerometer sensor embedded in the insoles inside shoes. It further focuses on the Z-axis of the data, showing how fall detection can be viewed with a pattern in the time series data, then data was processed meticulously to be able to study the patterns without noise. The CNN model implemented resulted in an average accuracy of 98%, which shows it was able to differentiate quite well between ‘Fall’ and ‘NotFall’ labelled occurrences. This shows the effectiveness of such a system, and how important it is to implement it. Everyone has a right to feel in control and safe to live a long healthy life.