Workpiece Identification and Localization Based on Monocular Vision

In this paper, based on monocular vision, a workpiece positioning method is proposed for solving the workpiece grabbing program on the planar workbench of industrial assembly line. When an object has large affine deformations, contour information based methods may fail to estimate the object position. To solve this problem, we first use affine-scale invariant feature transform (A-SIFT) method to find the coarse matched feature points between the measured workpiece and the planned one. Then, a criterion is established to evaluate the matching accuracy of these coarse matched feature points. Finally, we select a well-matched subset from the coarse matched ones to calculate displacement and rotation angles between the measured workpiece and the planned one. Experimental results show that this method handles well for various workpieces with different shapes and achieves high matching accuracy. Especially, it is also applicable for large affine changes under different industrial scenes.