A study of a supporting system for reviewing trial lessons using image processing technologies

We propose a review supporting system of trial lessons with image processing techniques. Candidates of a high/junior schoolteacher practice in the trial lesson. A subjective evaluation from the other candidates is used as a feedback of the lesson. However, the quality of the feedback result depends on the skill and experience of the evaluators. Review using a recorded video of the lesson can improve the skill of the candidate, however this method requires a long time. Therefore, we focus on an objective evaluation from a recorded video, in which the behavior is visualized in static images. The system provides the three processes, visualization of teacher movement, extraction of writing on blackboard, and blind spot alert. Our implemental approach is to calculate an optical flow indicating human movement, and to draw segment for visualization. An exclusive OR between previous and target frames is calculated for detecting a region of writing. The blind spot by the candidate is estimated with the human position and the region of the writing, and an alert is provided as visual information in the timing. The accuracy of the alert timing are about 80% of the precision and recall with the author visual judgment. In addition, some positive comments are obtained in a questionnaire result, and the availability is indicated in this paper.