Human Behavior Recognition by a Mobile Robot Following Human Subjects

Abstract

Our research is focused on the home healthcare support system for motor function impaired persons (MIPs) whose motor function should be closely monitored during either in-hospital or at-home training therapy process. Especially, for the at-home monitoring, the demand of which is increasing, not only close observation, but also accurate behavior recognition of daily living activity, as well as motor function evaluation, are necessary. In this study, such a system was established by developing a cost-effective, safe and easy to use mobile robot. With such a robotic monitoring system, the in-hospital time for most MIPs and the burden to therapists can be significantly decreased. In order to realize the robotic monitoring system, we proposed several algorithms to solve the difficulties arising from the mobile sensing for moving MIPs, and recognize several frequent daily living activities, including impaired walking. Concretely, algorithms to use both color images and depth images was proposed to improve the accuracy of MIPs measurement, and a Hidden Markov Model (HMM) was implemented to deal with the uncertainty on time sequence data and relate the state transitions over time for daily living activity recognition. Experiments have demonstrated promising results on joint trajectory measurement, and recognition of daily living activities.