Predicting Behaviors of Residents by Modeling Preceding Action Transition from Trajectories

Abstract

The smooth provision of support to residents by information display systems or robots will essentially require that their behaviors be appropriately grasped and that predictions be made that allow some margin for preparations. In this paper, we offer new perspectives by proposing a novel method to predict residents’ behaviors. The proposed method mainly consists of the following two phases: (1) to grasp the chains of residents’ potential actions from their trajectories, and then, (2) to identify the rules of association between residents’ behaviors, subject behavior to support and their last actions. In order to verify the performance of the proposedmethod in predicting residents’ behaviors, we have conducted experiments using two residents’ trajectories that have been tracked for around one year.