A system for an anticipative front human following robot

Abstract

Human following of mobile robots has been popularized in the scientific field of the service-oriented robots research. However, the behavior of such robots is passive in the most of cases. Compared to a passive following robot, a framework for an anticipative front following robot is proposed which is based on the technology of Kinect camera and the identification of human sensing and motion system. The real-time data is collected by Kinect and applied into the human walking model after median filtering. The human walking model is designed for identifying four basic human walking behaviors accurately. Then a human walking behavior is predicted using the optimized Kalman filtering algorithm proposed specially for the human walking model to predict the next behavior and provide more precise data for the system of robot control, which both could raise the success rate of accurate tracking. The system performance is assessed through experimental comparison of two specific human following behaviors: anticipative front following and passive front following. The results show that the success rate of predicting human walking behavior and following in front achieves 95.2%.