Abstract
Motion-picture analysis targeting the movements of people has been the subject of much research for some time. Some of this research has focused on the automatic video recording of lectures where the movement of the lecturer, the "lecture facilitator, " must be recognized and the recording cameras controlled on the basis of that movement. In a lecture archive system, the processing performed for recognizing the state of the lecturer consists of three stages. These are "lecturer-image recognition" to determine the position of various body components of the lecturer; "lecturer-basic-movement recognition" to determine the state of each component as obtained from lecturer-image recognition; and "lecture-state judgment" to determine the state of the lecturer according to the combination of component states as obtained from lecturer-basic-movement recognition. This paper reports on the results of experiments performed to compare several recognition techniques with the aim of establishing techniques for two of these stages, namely, lecturer-image recognition and lecturer-basic-movement recognition.
