Canny Edge Based Self-localization of a RoboCup Middle-sized League Robot

In the RoboCup middle-sized soccer league, autonomous robot players in the field are required to localize themselves in real time by utilizing the sensors mounted on themselves. There could be many different kinds of sensors that can serve as part of the solution to this task. Especially, vision sensors are one of the most promising because they are small, light-weight, inexpensive, and the most importantly, easy to be implemented and integrated as a real time subsystem in a mobile robot. Although the accuracy of vision sensors is subject to noise from camera configurations and environmental conditions, well-established image processing techniques are available and expected to overcome the difficulties. In this project we adopted an omni-directional vision sensor attached on top of a RoboCup middle-sized league robot (Corbenic) as the only available sensor for the self-localization purpose. The vision sensor has a perceptive range which is enough to cover the most of the soccer field regardless of robot’s locations. From the omni-vision camera, spherical images are constantly captured at the rate of 30 frames per second. A GPU (Graphics Processing Unit) on the robot’s main board is capable of carrying out the Canny edge detection algorithm at the same rate of the image updates. The goal of this project is to design and implement a real time subsystem capable of self-localizing a robot in the soccer field by utilizing the Canny edge information. We designed and implemented a localization method which takes Canny image data as input and provides a robot’s pose (2D position in the soccer field coordinate and robot’s orientation) as output. An experiment was conducted using omni-vision images captured in a real scale RoboCup soccer field in University of Ulm. The result showed that our method was highly feasible and expected to work even better with the help of estimation theories such as Kalman filtering and/or fusion with the other kind of sensors such as wheel encoders and polarization detectors.