Vision Calibration and Processing on a Humanoid Soccer Robot

In RoboCup, the problem of quickly and accurately processing visual data continues to pose a significant challenge. The Aldebaran Nao, currently used by the Standard Platform League, has two cameras for visual input, of which only one has been typically used. The integration of both cameras presents a new opportunity but also a challenge. While it is possible to obtain better information using both cameras, more cameras require more work to calibrate. We propose a novel camera calibration algorithm which automatically tunes a camera such that its color perceptions match those of another camera. Additionally, recent vision challenges introduced in RoboCup have necessitated the use of higher resolution images. We build on existing work in color based segmentation and present novel extensions to facilitate the move to higher resolution images, including memory optimizations, fast line and curve detection, and differentiation via robot pose based transformations. All work presented in this paper was successfully used by the UT Austin Villa Robot Soccer team, which secured 3rd place overall and 2nd place in the technical challenges at RoboCup 2010.