Cylindrical surface monocular reconstruction using virtual image plane method

A monocular vision system with straight line structured light is proposed to address cylindrical surface reconstruction problem. The vision system uses a virtual image plane camera model. The distortion is corrected by computing the virtual projection image on a virtual image plane. The projective line of a world point is determined by its virtual image point and the camera center, which can be localized by the proposed method. The structured light is calibrated from the virtual images of two stripes on two parallel planes. In accordance with the principle that the stripe of the straight line structured light on a cylindrical surface is an ellipse arc, the projection point's coordinates on the light plane are firstly calculated for each feature point of stripe from the virtual image. Then the ellipse is extracted from the arc points. Secondly the cylinder's diameter is acquired because it equals to the length of the ellipse's minor axis, then the cylinder is localized in 3D space. Finally, each projective line connected by the camera's optical center and the virtual image point is calculated. All intersections of the projection lines and the cylinder construct the point cloud of the cylindrical surface. Then the surface image can be reconstructed. In real experiments the accuracy of the vision system is close to that using binocular vision system. The effectiveness of the virtual image plane method is proved. © 2017, Science Press. All right reserved.