Research on Fall Detection Algorithm for the Elderly Living Alone Based on YOLO
Yanwei Yin, Liang Lei, Minghui Liang, Xiaobing Li, Yuanyuan He, Lanyao Qin · 2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT) · 2021
The 21st century is an era of rapid population aging, and accidental falls have gradually become the number one killer threatening the health of the elderly. Nowadays, the rapid development of computer vision provides a new solution for elderly fall detection. However, traditional machine learning methods have disadvantages such as cumbersome detection steps, poor real-time performance, bloated model deployment, and poor robustness in complex scenarios. This paper uses the YOLO series of algorithms to automatically extract features, and complete the end-to-end prediction of the target frame and category at one time. In this paper, two modified YOLOV4 and YOLOV5S networks are used as fall detection models. The network is classified and trained on a self-made fall data set, and tested in real scenarios to compare the performance indicators of the two models. Experiments show that the use of YOLOV5S can basically achieve real-time and accurate end-to-end detection and the model is lightweight and easy to deploy, and has good robustness in complex environments.