Research on the Improved YOLOv5 Algorithm Fall Monitoring System for Home Care for the Elderly
Wanchao Xu, Junchao Zhang, Xinyu Zhang, Fengxia Li, Jinyi Tao, Rongli Zhao · 2024
In view of the risk of fall for the elderly in the home care environment, a surveillance video display fall detection method of the improved YOLOv5 algorithm is improved, and the Transformer module that increases the human attention mechanism is combined with the CSPDarknet structure of the backbone part of the YOLOv5 algorithm. The system collects video data through the camera, trains and verifies the video data, and then judges the action and posture of the target person. It connects the WIFI module of Orange School to send the action and posture information to the guardian’s mobile phone to remind the guardian to timely treat the person who fell down, so as to improve the rescue efficiency. On the basis of improving the algorithm structure, the loss function was improved and conducted repeated training. The experimental results showed that the improved algorithm mAP improved by 2.8%, the $F 1$ score by 2.2%, the number of processing frames per second by 10% during inference, and achieved the fall detection accuracy of 98.6%.