On Non-Invasive Fall Detection Based on Multimodal Fusion
Huabei Nie, Dan He, Jianqiao Shen, Songchun Wang, Ani Dong · 2024
With the increasing aging of the population, the demand for fall monitoring for the elderly in homes and communities is growing. Traditional fall detection methods, such as singular computer vision or sensor techniques, often fail to efficiently recognize fall events due to certain limitations. This paper combines environmental factors and physiological signals from the human body among other modal information, and employs a multimodal learning approach for fall detection. We propose a multimodal fall detection method that integrates feature detection based on deep learning and feature computation based on human skeletal keypoints. Through this fusion strategy, the system can more accurately identify fall behavior and achieve efficient monitoring in various living environments. Experimental data shows that the fusion model has a higher accuracy rate and a lower false-positive rate than any single model.