SVM-based Identification and Un-calibrated Visual Servoing for Micro-manipulation

This paper presents an improved support vector machine (SVM) algorithm, which employs invariant moments-based edge extraction to obtain feature attribute. A heuristic attribute reduction algorithm based on rough set's discernible matrix is proposed to identify and classify micro-targets. To avoid the complicated calibration for intrinsic parameters of camera, an improved Broyden's method is proposed to estimate the image Jacobian matrix which employs Chebyshev polynomial to construct a cost function to approximate the optimization value. Finally, a visual controller is designed for a robotic micromanipulation system. The experiment results of micro-parts assembly show that the proposed methods and algorithms are effective and feasible.