Abstract
Deep learning technology has been widely used in edges. However, the limited storage and computing resources of mobile devices cannot meet the real-time requirements of deep neural network computing. In this paper, we propose a safer respiratory rate monitoring system using a three-tier architecture: robot layers, edge layers, and cloud layers. We decompose the task into a three-tier architecture of a lightweight network according to the computing resources of different devices, including mobile robots, edges, and clouds. We deploy feature extraction tasks, Spatio-temporal feature tracking tasks, and signal extraction and preprocessing tasks to robot layers, edge layers, and cloud layers, respectively. We have deployed this non-contact respiratory monitoring system in the Second Affiliated Hospital of the Anhui Medical University of China. Experimental results show that the proposed approach in this paper significantly outperforms other approaches. It is supported by the computation time cost of robot+edge+cloud architecture, which are 2.26 ms per frame, 27.48 ms per frame, 0.78 seconds for processing one-minute length respiratory signals, respectively. Furthermore, the computation time costs of using the proposed system to calculate the respiratory rate are less than that of edge+cloud architecture and cloud architecture.
Abstract (translated)
URL
https://arxiv.org/abs/2011.08482
