This paper described the tracking of robots and the ball for our RoboCup Small-Size team. We implemented several mechanism that make the system robust and fast, such as dynamic search windows, global search, color maps and robot identification. One important aspect was the careful design of the robot's markers, such that no combination of the markers from two different robots can be seen as a robot and such that there is some redundancy, for the case that one marker cannot be localized.
The FU-Fighters used the described tracking system in Amsterdam, where we won the European Championship 2000 and the improved system during the RoboCup'2000 competition in Melbourne, where we finished second, next to Big Red from Cornell University. The system delivered reliable information even in situations where other teams failed to localize the ball or the robots, e.g. when the ball was inside the wall's shadow or the referee was projecting a shadow on the field.
In the future, we plan to use a digital camera that can be connected
to the PC via an IEEE-1394 link.
This camera produces an uncompressed YUV 4:2:2 image with 640
480 pixels at a rate of 30Hz. The digital link, as well as the progressive scan should
increase the image quality and therefore simplify interpretation.
We are also investigating the possibility of adding an omni-directional camera to the robots for local vision. The image will initially be transferred via a wireless analog video link to an external PC, where it can be analyzed. We plan to apply the tracking principle to edges, e.g. the ones between the green field and the white walls to localize the robots. If this can be done with a low computational load, it would be possible to implement the local computer vision on a small, low power on-board computer, such as a PDA.