Robust and Efficient Embedded Vision for Aibo in Robocup

We present the embedded vision system used in the four-legged league which is one of the official leagues in Robocup competition. This vision system runs at frame rate in AIBO robots, while detecting colour coded objects and field line features. The detection of field line features is becoming necessary because RoboCup’s rules change year after year in order to push progress toward real soccer field where there are not characteristic coloured cues. Because four-legged league is quite dynamic, efficient and robust image processing techniques should be used. Furthermore, vision calibration should be as simple as possible because this process has to be repeated frequently. In order to choose the more suitable techniques and their combination, different methods are compared in terms of efficiency and robustness.