Supervisors: Louis Lettry, Alex Locher
The RoboCup Standard Platform League (SPL) is a robotics competition of soccer playing NAO robots. Former vision projects allow the robots of the ETH Zurich team to properly detect the ball and estimate the player position based on the field lines. In this semester project, an approach to detect players using the random forest classifier is presented. Stable detection of robot players is needed to avoid bumping into other players and to improve awareness of the current game situation. In a first step, the performance of random forests trained to detect different parts of the NAO robot was evaluated. To improve detection reliability a multi-step detection was developed and implemented that combines intelligent resampling to detect multiple robot parts and correlate the findings. The results show that a good robot detection can be achieved with this approach.