"Image Processing for Flood Warning"
Behind my house flows the Little Creek, which drains rainwater a large portion of Chapel Hill by way of the now-infamous Bolin Creek and Booker Creek. These upstream tributaries of Little Creek regularly flood and frequently damage the Eastgate shopping center and camelot village apartments.
I recently set up a webcam in my backyard that periodicly stores a picture of the creek area and immediately makes it available on the web. (http://www.telltronics.org/creekcam - be gentle, its a rather wimpy DSL line)
I'm planning very soon to place a painted and/or illuminated reference marker in the field of view of the camera, as a visual aid in estimating the water level from the pictures.
The software project: analyze image data from the camera, and estimate the water level, both in daylight and at night. From that start, the software could add various other features: Generate web pages that contain the image, the numerical estimate of water level, and some kind of confidence metric in the estimated level. Log the estimated level over time. Send email when the level exceeds a certain threshold. Integrate camera, video capture, analysis software, and web server into a standalone turnkey internet-connected appliance.
Winter rains usually elevate the creek to "interesting" levels; a suite of test images could be gathered then. Also, I believe that mockup flood scenes could be be set up relatively easily for collecting test images.
In order to be useful to me, the software developed would have to run on Linux and be entirely Open Source. Beyond that, implementation language and libraries/toolkits are relatively unimportant.