Abstract. This paper describes a two-tiered approach to the self-localization problem for soccer playing robots using generic off-the-shelf color cameras. The solution consists of two layers; the top layer is a global search assuming zero knowledge, and the bottom layer is a local search, assuming a relatively good estimation of the position and orientation of the robot. The global search generally yields multiple candidate positions and orientations, which can be tracked, and assigned a confidence level using the local search and/or historic information.