B-Human 2016 - Robust Approaches for Perception and State Estimation Under More Natural Conditions

In 2015 and 2016, the RoboCup Standard Platform League’s major rule changes were mostly concerned with the appearance of important game elements, changing them towards a setup that is more similar to normal football games, for instance a black and white ball and white goals. Furthermore, the 2016 Outdoor Competition was held in a glass hall and thus under natural lighting conditions. These changes rendered many previously established approaches for perception and state estimation useless. In this paper, we present multiple approaches to cope with these challenges, i. e. a color classification for natural lighting conditions, an approach to detect black and white balls, and a self-localization that relies on complex field features that are based on field lines. This combination of perception and state estimation approaches enabled our robots to preserve their high performance in this more challenging new environment and significantly contributed to our success at RoboCup 2016.