A Fall Monitoring System Based On Embedded System
Ziyu Guo, Qing Xiang, Limin Cai · 2023
In today's increasingly severe aging age, the safety requirements of the elderly living alone both in indoor and outdoor activities are increasing, in response of this social phenomenon, this paper completes a lightweight embedded fall detection system by combining embedded technology, network communication technology and the behavior recognition module of the target detection algorithm using deep learning. This study used 534 images containing fallen objects as a dataset, based on the YOLOv5 (You Only Look Once) algorithm model[1]. The model was lightweighted, and the model network structure was improved by densely connected networks. In the experimental comparison with the original model, the new model achieved the best performance with fewer iterations, and the recall rate of 97.74%, which was 1% higher than the original model; the best accuracy of 97.09%, which was 0.96% higher than the original model; and the mAP value of the improved network of 0.962.