A low-power fall detection method based on optimized TBM and RNN

Abstract

Falls are one of the most common accidental injuries among elderly, which would cause injuries such as bruises and fracture. The fall detection system can reduce injuries by promptly notifying health care providers. Existing methods mainly focus on performance optimization of fall detection algorithm. However, it is unsuitable for the application of wearable devices because it increases the complexity of the algorithm and the power consumption of wearable devices. In this paper, a fall detection technology based on wearable device (placed at the waist) and cloud platform is proposed to minimize the fall-related injuries. Low computational complexity Threshold Based Method is implemented i n wearable device to save power. Furthermore, in the TBM phase, computationally simple acceleration correlation features are selected and the thresholds are optimized to decrease the uploaded data. On the cloud platform, the RNN algorithm with high performance is deployed to recognize suspected fall. Finally, the performance of fall detection system is improved to sensitivity (98.26%) and specificity (99.21%). In conclusion, the algorithm performance and power consumption can be balanced perfectly in this paper.