Proven Techniques for Robust Visual Servo Control

Abstract

This chapter discusses a variety of techniques that can significantly improve the robustness of visual servoing systems. Unlike other disciplines that address robustness issues such as robust control, we do not adopt a formal notion of robustness but rather concentrate on two phenomena which are typically the reason for a failure of vision systems: unexpected occlusion and infavorable illumination conditions. We discuss a variety of image processing techniques that ensure stable measurements of image feature parameters despite these difficulties. In particular, we will cover feature extraction by Hough-Transforms, color segmentation, model-based occlusion prediction, and multisensory servoing. Most of the techniques proposed are well known, but have so far been rarely employed for visual servoing tasks because their computational complexity seemed prohibitive for real-time systems. Therefore an important focus of the chapter is on efficient implementation. The robustness of the techniques is highlighted by a variety of examples taken from space robotics and medical robot applications.