Smart Fall Detection for Elderly with IoT-Based Ground Vibration Signal Fusion

Ageing makes an elderly person vulnerable from loss of control as well as occasional to frequent loss of consciousness; as a result, detection of the fall of an elderly person received considerable attention. Fast detection of the fall and prompt notification can increase the probability of survive at a significant rate. It is observed that while existing vision surveillance solutions hinder the privacy of a person, wearable devices are body worn 24X7. In this paper, an IoT enabled sensor fusion scheme is proposed for elderly fall detection in an indoor environment. An important advantage of the proposed idea is that the privacy of the person is ensured intact. In the proposed approach, an energy detector comparing background and foreground ground vibration is deployed to continuously detect a footfall. When a footfall is detected by the ground vibration sensing module, a camera records the footage of the movement and extracts 33 landmarks each, for fall as well as other poses of a human. The landmarks are then uploaded to the cloud for estimating the actual pose of the human in the scene. Note, analysis of popular algorithms for binary classification of fall versus other poses indicated that a light gradient boosting machine learning algorithm outperforms those in comparison. Performance of the proposed approach is evaluated in a real ambience, wherein, the floor movement detector provides 96.3% accuracy with very low false alarm rate; also, the fall detection module (using the pose estimate) yields an accuracy of 99.3%.