Fall Detection Method Based on Improved YOLOX Network

Abstract

The global aging problem is deepening, and the safety care of the elderly will receive wide attention from all walks of life, among which falls are the primary factor leading to disability and death of the elderly. To address the shortcomings of traditional video-based detection methods, which cannot balance detection speed and accuracy, this paper proposes a fall detection method based on YOLOX network, which conducts training tests on three common human actions: standing, falling and sitting, and improves the accuracy of target detection by improving the structure of YOLOX network and adding two attention models to compare experiments. The effect of the improved model is compared with the original network to demonstrate that the proposed detection algorithm has higher accuracy.