Robust Perception of an Interaction Partner Using Depth Information

Abstract

Abstract—Social interactive robots require sophisticated per-ception abilities to behave and interact in a natural way. The proper perception of their human interaction partners plays a crucial role. The reduction of the false positive rate for human detection is very important for increasing the natural interaction abilities. This paper presents a combined method using RGB data as well as depth information to find humans in the robot’s surrounding. To track a person over time a Kalman filter is applied, which also reduces the processing time. Furthermore, a head pose estimation on the basis of Support Vector Machines is integrated, which can be used to perceive nonverbal expressive cues like nodding. The proposed method is tested in various experiments. Keywords-social robots; perception; human-robot interac-tion; I.