A statistical method for color object detection and application to localization

In our goal to design an autonomous vehicle, we face the problem of robot localization. Vision and landmarks bring some natural solutions to the problem. The next difficulty lies in the recognition of the landmarks. The paper proposes an approach for color object detection, based on normalized color vector clustering using relaxation methods, followed by a multidimensional discriminative analysis. We then show how to simplify the problem and present an application to localization using a three-color fiducial. The method is robust to light intensity variations.