I think the first thing that comes to mind is the amount of needless repetitive motion on the operator's part. The camera shutters should be triggered automatically, shortly after the frame settles down on the book. You could do the job mechanically, optically, or with an accelerometer for that matter. If the operator didn't have to let go of the apparatus to work the camera shutters, it would easily double the throughput.
I'd probably aim for something close to the ( http://www.youtube.com/watch?v=EgiPaSOeP2w ) Booksnap digitizer, but without the glass frame to hold down the pages. Given a pair of good high-res cameras, the page curvature can be taken out in software. Further, once you're doing any sort of image postprocessing at all, it wouldn't be hard to sense when the user's hands have been withdrawn after turning each page. As soon as a couple of frames have been taken without any significant motion, the scanner can keep those images and signal the operator to turn the page.
Those all sound like wonderful ideas, but ones that would be very complicated to implement. The design they show costs around $250, the price of the two consumer grade cameras and the plexi + mounting screws. It is something a student could put together and use in a dorm room with relative ease.
Since locked down, limited firmwares on all consumer grade cameras prevent any kind of direct computer control, and webcams top out at a grainy 3MP with poor optics, you'll have to greatly increase the cost as well as complexity of the system if you want to do it all in software. The Booksnap scanner you link uses cameras that start at 500-600. And you'd have to write the software of course.
I'd probably aim for something close to the ( http://www.youtube.com/watch?v=EgiPaSOeP2w ) Booksnap digitizer, but without the glass frame to hold down the pages. Given a pair of good high-res cameras, the page curvature can be taken out in software. Further, once you're doing any sort of image postprocessing at all, it wouldn't be hard to sense when the user's hands have been withdrawn after turning each page. As soon as a couple of frames have been taken without any significant motion, the scanner can keep those images and signal the operator to turn the page.
Google, according to http://www.boingboing.net/2009/05/03/how-googles-book-sca.ht... , apparently gets their page-curvature feedback from infrared cameras, but I don't think most home users would need to get that fancy.