Visual Robot Detection and Free Space Recognition for the Standard Platform League

Playing soccer is one of the standardized problems supplied by the RoboCup project. Soccer playing robots acquire all information on the condition of the environment from their sensor data. Especially the preparation of visual sensor data is essential for the decisions of the robot during play. Besides the recognition of the ball and the goals, anticipating free space, as well as the position and distance of obstacles, here in particular other players, is an important issue. The knowledge where the field is occluded and where it is free to go to is utilizable for planning collision-free paths. Furthermore, such information aids to refine reactive behavioral strategies and is conducive to team play capabilities. The aim of this work is to extract information both on the free space in the field, as well as on the position of other robots from the visual sensor data. Two approaches for free space detection are presented. The strategy of the first one is to scan images for the beginning of the green field starting at the upper border. In the other approach, images are traversed bottom-up and green regions, reachable by walking straight on, are considered. For the robot detection a neural network algorithm is implemented. On the basis of color segmented images a feed-forward neural network is trained to discriminate between robots and non-robots. The presented algorithm initially extracts image regions which potentially depict robots and prepares them for classification. Preparation comprises calculation of color histograms as well as linear interpolation in order to obtain network inputs of a specific size. After classification by the neural network, a position hypothesis is generated. The algorithms are integrated into the B-Human Software Framework 2009 and evaluated in terms of speed and accuracy.