Fall Detection for Elderly Person Monitoring using Wearable Inertial Sensors and Locality Sensitive Hashing

Falling represents the main threat for elderly people. An early detection mechanism is crucial to assist the person and prevent further injuries from concussion to underlying pain from fractures. In this paper, we propose a new model for fall detection to improve the detection accuracy and to reduce the detection delay, while considering detection processing in the sensor to reduce the energy consumption by wireless transmission of Inertial data from sensor to processing unit. Our proposed approach is based on change detection in the slope of angular velocity, which enhances the sensitivity and drastically reduces the detection delay. After the detection of change, the local sensitivity hashing is applied to detect change in the amplitude of accelerometer signal. We applied our multimodal system on publicly available data set built for fall detection during daily activities. The proposed approach uses only data from 3D gyroscope and 3D accelerometer acquired by inertial sensor from different positions: ankle, wrist, waist, pocket, neck and forehand. Our experiment results show that our proposed model achieves high detection accuracy with low false alarms when compared to existing work in the literature. Furthermore, we compare our approach with existing work in terms of computational complexity and detection delay to prove that it reduces the energy consumption and achieve better performance than existing models.