The capture of whiteboard images, while required for systems such as ReBoard, is not particularly interesting these days from a research perspective. What’s more interesting (and what’s interesting to us about ReBoard) is how this captured information can be used.
We’ve been using video cameras as network-accessible Axis digital still cameras for ReBoard, and they produce reasonable, but by no means great, images. They work pretty well in our offices, where they can be mounted on the opposite wall, and calibrated to image the whiteboard. We run the captured images through a process that corrects the distortion and extracts the whiteboard region, and then shove the images (both the original and the distorted one) into a database. Works great.
But what happens when the whiteboard moves?
That is, what if the whiteboard isn’t attached to the wall, but can be moved around? What if you can’t attach a camera to the opposite wall because there isn’t an opposite wall?
One class of solutions that we’re looking at to make this tool useful in a broader range of environments is to look at how a digital camera (with or without a cell phone attached) could be used to collect images. Evernote, for example, make capturing content easy, and can even perform limited handwriting recognition on the captured image. But then the data are stuck in Evernote, and that limits our ability to explore novel ways of retrieving and reusing the data.
Another promising solution is a camera like the Casio Exilim that has a whiteboard detection mode. It identifies the board in the frame (optionally with some feedback from the user) and then corrects the image to compensate for off-axis shots. Once uploaded, the whiteboard images could be selected and added to ReBoard automatically. Of course having on-camera whiteboard detection is not necessary, as all images could be scanned and processed by our algorithms , but giving the user feedback about the quality of capture at the time the image is captured (while there is an opportunity to re-take the shot) should be useful.
The issue of which whiteboard was captured still needs to be addressed, of course, and we would lose our nice mixed-initiative approach where the system captures content automatically even if the user doesn’t take an explicit snapshot. We believe this is an important usability feature of the system, because it removes some of the burden of capture, but on the other hand it requires a level of infrastructure that may be impractical (and expensive) to deploy in many circumstances.
Thus it should be interesting to see just how much automation is required to make the system sufficiently easy to use to allow it to be useful.