Identification of friendly and deictic gestures in a sequence using upper body skeletal information

摘要

Human-robot interaction has become a popular topic in the domestic environment owing to the increase in elderly population. In order to achieve a more natural interaction process, robots are expected have human-like interaction capabilities. Human-like behavior should extend beyond the interaction process to perceive human intention. The robots are required to have intelligence to interpret and distinguish a human user's instructions as well as friendly behavior before making decisions regarding interaction. This paper presents a model that can identify a sequence of gestures without the interference of voice commands. The system extracts and analyzes upper body skeletal information for a period of time in order to identify friendly and deictic gestures separately, before initiating an interaction on user demand. Ability of the robot to perceive friendly or calling gestures alongside deictic gestures make the task of commanding the robot easier and nonverbal. This behavior enables interaction process even in noisy environments as well as at relatively longer distances. To achieve this behavior, the system is equipped with separate units to extract and analyze information, estimate 3 types of gestures in series and finally to take decisions regarding whether to interact with the user. The system is implemented and tested in an artificially created domestic environment. Results of the conducted experiment are used to verify the behavior of the proposed system.