A Fall Detection System for the Elderly Based on Head Speed and Figure Variations

Abstract

The invention concerns the development of a fall detection system for the elderly, adults and children. Our system relies exclusively on surveillance cameras for smooth, unobtrusive daily monitoring. Our approach combines human silhouette and head speed for fall detection. The head is detected from key points and the silhouette mask extracted at pixel level using a pre-trained neural network. Head velocity is calculated by detecting the head centroid and measuring the distance traveled by this point during activity. This advanced method guarantees accurate and efficient classification of falls, and can be used to detect head injuries caused by impact with the ground at high speed.