Automatic Evaluation of Learning Behaviors for Online Lectures by OpenPose

To prevent that learners do something unrelated to an online lecture, we proposed an automatic evaluation method of learning behavior. This method extracted the pose of learners and a teacher by OpenPose, constructed a model that estimate to a teacher’s pose at j sec in online lecture from a learner’s one, and evaluated by measuring the difference between estimated and actual values. The result of an experiment showed this could detect the unrelated actions about 90%.