Utilizing the structure of field lines for efficient soccer robot localization

Self-localization on the field is one of the key perceptual tasks that a soccer robot, e.g. in the RoboCup competitions, must solve. This problem becomes harder, as the rules in RoboCup more and more discourage a solely color-based orientation on the field. While the field size increases, field boundary markers and goals become smaller and less colorful. For robust game play, robots, therefore, need to maintain a probabilistic pose estimate and rely on more subtle environmental clues. Field lines are particularly interesting features, because they are hardly ever completely occluded and observing them significantly reduces the number of possible poses on the field. In this work, we present a method for line-based self-localization on a soccer field. Unlike previous work, our method first recovers a line structure graph from the image. From the graph, we can then easily derive features such as lines and corners. Finally, we describe optimizations for efficient use of the derived features in a particle filter. The method described in this article is used regularly on our humanoid soccer robots, which won the RoboCup TeenSize competitions in the years 2009–2011.