Supervisors: Dr. Ajad Chhatkuli, Jan-Nico Zaech and Prof. Luc Van Gool
The "Robot Soccer World Cup", also known as RoboCup, is a student league where fully autonomous humanoid robots compete against each other in teams of five. One of the main challenges for the robots is to localize themselves which is done through extrinsic calibration of the cameras during the game. For the exact and efficient self-localization from the camera pose, it is important to determine the offsets between the camera's orientation and the robot's head orientation. The state-of-the-art method computes the orientation angle offsets in a controlled setup by marking lines and points with human input. We provide an alternative method that requires minimal user inputs by warping the images such that the field lines suffer from minimal perspective changes. We then detect lines and filter out false lines using various heuristics in order to establish accurate correspondences to the 3D field coordinates. Our method performs the calculations in a matter of seconds compared to 10 minutes from the currently used methods and much more accurately than the feature correspondence based methods.