Fall detection using mixtures of convolutional neural networks

Falls may happen to everyone; however, with geriatrics, this factor is one of their primary concerns as it might cause detrimental effects on their health or perhaps unintentional death if the case is terrible. To tackle this problem, many scientists have undertaken a considerable amount of research to create a fall detection system. This paper presents a fall detection architecture using a Mixture of Experts (MoE) and CNN3D models on a large public dataset called UP-Fall Detection. Furthermore, we also utilize the data augmentation approach to tackle imbalanced problems in this dataset. Our methods can gain a significant result with 99.67% in weighted average F1 score, which is necessary to build a fall detection system. Model and code are available at https://github.com/hoangNguyen210/Fall-Detection-Research-2.