Automatic and easy creation of learning friendly conferences videos A practical example of computer-vision applied to online-learning

摘要

Nowadays, a lot of conferences, meetings, symposiums, etc. provide video reviews of the talks that have been taking place. Using a professional video editing software, the creation process of those videos can be painful and time-consuming. Using computer visions techniques like background/foreground subtraction, particle filtering and face detection for discovering interesting parts of the talk we can leverage this difficulty to a quasi null factor, meanwhile speeding up the process. While usual approaches use separate audio/video/presentation files, we propose to recover everything directly from a High Definition video, stream of the talk. The produced meta-data and video allows seamless integration with Scene Knowledge, a video reviewing on-line system developed by NTT.