Real-time fall detection system based on postural changes

To enable real-time detection of falls among elderly individuals, this paper presents a fall detection system based on the dynamic changes in human body posture. This system employs inertial sensors to collect posture data from elderly individuals. After preprocessing and feature extraction, the system utilizes the ReliefF algorithm to determine feature weights. Based on this, it establishes a fall risk index. Subsequently, four Support Vector Machine (SVM) models with different kernel functions are trained. The predictions from these models are then combined using a voting mechanism and integrated with genetic algorithms, resulting in a multi-kernel SVM ensemble model. Finally, a comprehensive algorithmic model for fall detection is established. Through experimental testing, this comprehensive algorithmic model achieves an accuracy rate of 99.3%, thereby confirming the feasibility of the system.