Robust adaptive vision for robot soccer

Robot soccer requires vision to be fast, accurate and reliable for a team to perform well. To win competitions, the vision system must work every day under all contest conditions. In practice, the field conditions for robot soccer can vary enormously from field to field, and even from day to day throughout a contest. This paper presents a system that can adapt to rapidly varying light and colouring conditions, while maintaining speed and accuracy. Quantitative results of performance are presented, and applications outside the robot soccer domain illustrated.