Wavelet Entropy and Support Vector Machine Multi-Class Fall Detection Based on IMU Wearable Sensors Data

Fall detection is especially important for seniors because falls are one of the main health dangers for the elderly in public health care. Elderly people are particularly vulnerable to falls, which can result in serious injuries, especially if the person who falls remains on the ground for an extended period without help. In this paper, we proposed a fall detection system-based Inertial Measurement Unit (IMU) sensor to improve the detection method. Wavelet entropy for feature extraction and a directed acyclic graph support vector machine (DAG-SVM) classifier for multi-class classification are then applied. The proposed system is evaluated on a ready-made dataset based on multi-wearable sensors; data come from multi-sensors chest, thigh, and waist IMU datasets. The multi-class classification approach splits the elderly activities into three phases: pre-fall, impact, and post-fall. Experimental results, relying on using a single accuracy in fall detection were compared with multiple sensors and achieved 71.66%, 71.64%, and 97.79%, with different sensor locations. The results showed that the proposed model-based multi-wearable sensors demonstrated high accuracy compared to the model-based single sensor.