Monocular-Vision and CAD Model based Online Roll-Axis Registration

This study proposes a single-axis registration system based on monocular vision and CAD modelling without employing three-dimensional scanning or reconstruction techniques. The developed system uses dynamic modelling and image projection methods to estimate the single-axis rotation angle of an inspected workpiece. The estimated rotation angle of the inspected workpiece is achieved using the pose of the virtual model. Dynamic modelling significantly reduces both the data storage and computing time required to match the virtual model. Through experiments, we found that the average root-mean-square (RMS) error between the true centre location of special features within the inspected workpiece and those constructed within the virtual model was 2.19 pixels, and the RMS overlap rate was 96.44%. Thus, a monocular vision three-dimensional workpiece registration system with a dynamic model is effectively realised.