Kinect-based Behavior Measurement in Group Discussion

Abstract

Group discussion is a typical kind of learning scenarios. Although there have been various ICT-assisted approaches to measure the learning behaviors in classroom, few of them were proposed for group discussion. Since the student in discussion may sit around with unfixed positions, it is difficult to observe their behavior with fixed video cameras. In this paper, we propose a novel measurement scheme, which utilizes the depth sensor installed in the center of desk. In our scheme, four students are arranged in one group and sit around the desks, and two Kinect sensors are deployed to detect their body behaviors. To provide good vision for students, the Kinect sensors are put close to the horizontal plane of desks, resulting in low-angle shots of body gestures. We develop the algorithm to correct the measured skeleton data, and detect the facial direction of each group member, which enable us to infer the role of each member in the discussion, i.e. speaker or listener. Experiment results show that, this scheme can detect most of the body behaviors in group discussion, and can be utilized to quantify the student's proactivity and participation in the group.