Implement a simple recognition system for objects on the ground.

In this year's RoboCup@Home navigation challenge we will face the task to navigate through an environment where there are up to three dynamic obstacles. These obstacles will be rather small and lie on the floor. They will be too small to detect them with the laser scanner. Therefore a vision-based approach is needed to detect the objects and recognize them as an obstacle.

The idea is that we will have a rather homogeneous floor (the brown floor from Bremen comes to my mind) while the objects will have a distinct color from this. This way we can use the front camera and look for discrepancies in the floor color and assume these to be objects. If we then take the nearest point we can use the existing relative ball position model to calculate the position of the obstacle. This can then be merged into the obstacle laser as yet another "virtual" obstacle and thus the navigator will automagically circumvent the obstacle.

If the assumption that colors for floor and foreground objects can be rather easily detected (which seems likely) much of the old software can be re-used. It will basically be yet another incarnation of the front vision.