Heyall.

[problem background]
Myself and a gang of buddies have just started taking a university course wherein we are expected to build a robot (or similar device) from scratch, using whatever sensors, microprocessors, wiring and mechanical thingamajigs we desire (within reason). We are currently discussing what kind of robot to build. There are a number of "standard choices", and then there's our own idea.

We're considering trying to build a mobile ballistic launcher - essentially an autonomous, moving cannon. It would find a target (or number of targets), stop, aim and fire. If it hits, it moves to the next target, otherwise it adjusts its aim and fires again.
[/problem background]

The problem is that unlike other projects (which use lasers), this robot would need to analyze its misses in order to compensate. This means figuring out if the shot went high, low, left or right and how much. Disregarding that problem, we could simply use IR beacons as targets which go out when hit (IR sensors are easy to use), but because of the miss issue we will need to use regular cameras.

That means image recognitioning.

The processors we have access to can be programmed using C compilers available to us, which means we would obviously code the image recognitioning in that language. The environment will consist of 3 types of monocolored (besides shadows) surfaces: white walls, undamaged targets (black balloons?) and walls damaged by missed shots ("bulletholes" - probably paint). There would also be a chaotic background visible above the environment walls where people will be standing with cameras and stuff looking at the action.

Question: How hard is it to code an image recognitioning routine in C (no C++, I'm afraid) that is sufficiently sophisticated to distinguish monocolored targets and bulletholes from the mostly white background (avoiding walls will be accomplished using laser range finders)? How much processing power does it take? Are there any decent resources, algorithms or sample programs you can reference?

Thanks in advance.