Image based visual servoing from hybrid projected features

Abstract

in this paper, we address the issue of hybrid visual servoing, which improve the performance of robotics system. Different from most of existing hybrid methods combining image based(IBVS) and position based(PBVS) visual servoing methods, we consider directly image features in error function, which is a robust IBVS method. In the framework of unified projection model, we select six proper features to control full-DOF motion. Invariants to rotational motion are computed with spherical projection model and used to control the translational motion, while invariants to translation motion are computed with perspective projection model and used to control the rotational motion. These hybrid projected features ensure good decoupled control between translational and rotational motions. Furtherly, two virtual moments calculated with SVM regression model are proposed to control the rotational motions around the x-axis and y-axis, which ensure improved decoupled properties. Simulation results are provided to illustrate the effectiveness and optimal performance of the proposed method.