I used to work for a medical school where my primary job was automating the information capture (video and slides) of lectures. We had numerous lectures going on every day and only one or two people to post-produce the videos. One of the big problems we faced was the large variety of inputs: cameras, Powerpoints, writing surfaces, DVDs, etc. Professors couldn't be bothered to standardize on a single presentation format and would sometimes wait days after initially presenting to provide the slides for us to include with the captured content. I started working on a solution that would hack into the VGA signal feeding the projector and capture stills on a regular basis so at the end of the class, we'd have both the video AND the contents of the presentation. The problem with that was that it wouldn't capture full motion video, like when instructors displayed video through the projector, or had powerpoint slides with video embedded in them. My next idea was to capture images of the VGA input at 30 fps, analyze the difference between the current frame and the previous few, and make a determination if a region of the image was displaying full-motion video (if so, I'd send that region to an encoder, then subsequently overlay it on top of the high-resolution stills when displaying it to the students online). I took another job before finishing work on that idea, and since then the medical school purchased MediaSite Live from Sonic Foundry, which I think does a pretty decent job (not as cool as my idea, but workable).