그래프 합성곱 오토인코더 기반 고령자 이상 행동 인식

The increase in the elderly population living alone is considered the most serious problem. The elderly people living alone are difficult to cope with emergencies situations properly, leading to lonely deaths. Although various monitoring systems for emergency situations at home have been developed, some limitations still exist in identifying abnormal behaviors of the elderly people because of frequent equipment malfunctions. To address this situation, we use ETRI-Activity3D recorded in locations that mimic actual living spaces focused on the daily activities of the elderly people. In this study, we propose a methodology to detect abnormal actions of elderly people using a graph convolutional autoencoder. By capturing the spatial temporal feature of each action through graph convolution operation, the proposed method outperforms other existing methods in detecting abnormal actions. We hope that our work can help elderly people in real emergency situations.