Human activity recognition via motion and vision data fusion

Automated recognition of human daily activities is very important for human-robot interaction (HRI) in assisted living systems. We propose a Bayesian framework to integrate motion sensor observations and the location information from a vision system for human daily activity recognition. Two problems are studied in this paper: enhancing activity recognition through the fusion of two channels of information and learning the environment through the activity distribution map. The entropy associated with human activity recognition is adopted as an evaluation metric in both problems. The simulation results demonstrate the feasibility of the proposed methods.