The Internet-of-Things based Fall Detection Using Fusion Feature
Tuan-Linh Nguyen, Tuan-Anh Le, Cuong Pham · 2018
Many approaches to fall detection based on computer vision or a single sensor might struggle with the posture variation of falls, for example, the discrimination of fall and fall-like activities such as lying down. In this paper, we propose a method and system for fall detection based on the combination of features extracted from multiple sensors employed in the wearable device. Our proposed fall detection method comprises of four steps: data pre-processing, segmentation & event detection, feature extraction & fusion, and pattern recognition. In addition, the system exploits the Internet-of-Things (IoT) toward for energy efficiency on sensor nodes and real-time implementation. The proposed method is verified through an empirical experiment with the dataset collected from 26 users wearing the device and simulating 8 types of fall and 8 fall-like activities including unknown activities. The results demonstrate that the falls can be detected with 87% accuracy with a single sensor and up to 94% precision and recall with fusion features extracted and combined from multiple sensors under 10-fold cross validation evaluation protocol. These results are really promising for IoT-based situated applications for assistance of the elderly.