Vision attentive robot for elderly room

Abstract

With people's busy schedules, elderly people have to stay alone in their houses in the daytime. There are a large number of accidents have happened to elderly people when they are alone at home. It is crucial to have a monitoring system to identify the potential hazards for the protection of elders to address this risk. In this research, we propose a method to identify postural behaviors, walking abnormalities, and falling situations using the skeleton data obtained from the Microsoft Kinect camera. In this paper, we discuss the identification of the falling of an older person. For that, we used an LSTM model, and the features of the model are velocities of angles and joints of the skeleton. This system achieved a validation accuracy of 88.34%, and it offers a promising solution for keeping an eye on and recognizing potential dangers for elderly people.