High-speed communication networks and multimedia processing techniques enable us to attend a distant lecture. When the lecturing scenes are taken, a camera-person controls the camera in order to capture suitable scenes; otherwise, the camera is set to capture the same position at all times. Since the scene sent from a fixed camera is unable to give us the feeling of being at the live lecture, we would like to pan or tilt the camera to follow the teacher's action. Therefore, a mechanism to control a camera automatically is necessary. We describe experiments on an automatic camera system for taking lecture scenes. Computer vision techniques are used to implement the system. We have categorized the teacher's actions into three states. The first is a state explaining without a blackboard; the second is one writing on a blackboard or explaining with it; the last is one moving in front of a blackboard. We assumed that the teacher is in one of the three states in a lecturing scene, and have examined how to control a camera in each state. We have implemented two prototypes: one is a prototype following a teacher; the other is a prototype estimating the teacher's posture.
展开▼
