An Evolutionary CAD model-based Pose Measurement Method for Industrial Parts based on Monocular Vision

Abstract

We propose a 6-DOF pose measurement method for industrial parts with complex shape based on monocular vision. According to the CAD file information, the 3D model of an industrial part is established. Then, an offline template library is obtained by the 3D model under different observation views, to reduce the actual online measurement time. The similarity function between the image and the template is established by a Canny-based improved Chamfer distance matching algorithm. The Chamfer distance image is divided into four layers by using the direction angles of the edge gradient, to improve the sensitivity of the matching function. Genetic algorithm (GA) is used to search for the optimal matching result, which combined with the hill-climbing method to make the searching process converge quickly. The experimental results show that our proposed method can measure the targets with known complex shapes in a 3D working environment, with the position error is within 2mm and the rotation error is within 2°. For dynamic parts, our proposed method can achieve fast matching, and the matching is applicable to different dynamic target parts, the model matching is only related to the shape of the part.