Visual framing feedback for desktop video conferencing

Desktop video conferencing participants are often poorly positioned in their outgoing video - unaware of their appearance on camera. We combine face detection, feature tracking and motion detection for automatic real-time detection of poorly framed participants and subsequently provide framing feedback by compositing their incoming and outgoing video streams. Our solution provides participants visual feedback only while they have framing problems. Otherwise the display shows only the remote participants, allowing users to focus fully on the conference. We analyze the system components and describe a user study that found advantages of our approach over the existing mirror window solution.