Robust visual servoing with spheres projected in cameras obeying the unified model

Visual servoing (VS) is an automatic control technique which uses the feedback of visual information to guide the robot motion to converge to the desired location. This paper proposes new moment-based features for robust VS with spheres, especially in cameras obeying the unified model. Specifically, we introduce a virtual plane to reload contours of image projections of spheres, in the process of which image noises will however be greatly magnified and induce large converge error. To solve this, new moment-based features are proposed and extracted on this virtual plane to enhance the robustness of VS to calibration errors and image noises. Simulation results verify the effectiveness of the proposed scheme and demonstrate the high robustness of these set of new features to image noises. The altering trends of converge errors versus increasing noise levels are randomly tested to present the obvious advantage of new moment-based features over the usual ones. At last, a path-planning based VS application is performed with fisheye views of some spheres, benefiting from the robustness and an interpolation-based tracking scheme.