Results and Future Work

We were able to localize the robots and track the ball and goals with a rate of 25 fps. The localization was correct in about 95 % of all cases. However, the precision of localization must be improved further. Here, the primary problem is the distance mapping of a point in the real world to a pixel in the image. This mapping is imprecise due to misalignments of the optical system that are enforced by robot movements. To cope with the above problem, we plan to develop an automatic calibration method, which autonomously adapts to changing mapping situations.
Another problem is color classification. Although our color classification was very robust and successful compared with other teams, the fact that the user has to interact with the program to specify color classes infringes the idea of an autonomous system. Thus, we will focus our research on finding auto-configurative methods. Furthermore, we want to develop a vision system that is able to fuse many different techniques for localization, detection and tracking of objects. On a higher level such a vision system will allow to define different visual behaviors, enabling the user to specify how active vision and visual attention is used in different situations. We want the system to be flexible, such that RoboCup is just one application the system can cope with.