Binocular Stereo Vision System For The 3D Position And Orientation Of Workpiece

In robot manufacturing is difficult to measure the position and pos of large components with complex surface,complex curves,because it is usually inaccurate when installing large components whit fixture.Thus the robot program can not be reused.The conventional approach is to re-teach the robot or manually measure the transformation matrix,which restricts the speed of pipeline manufacturing.In order to solve the problem of measure the position and pos of large components,a binocular vision system is append to robot manufacturing system.Binocular vision system make it possible to quickly locate the position and pose of the workpiece.The robot can automatically correct trajectory according to the position and pose information of workpiece.In this paper,we first describe the binocular vision system structure.Then we establish the mathematical model of binocular vision system,relationship between the camera and the robot.To compute the parameters of mathematical model,a series of experiments were performed.After calibrat the parameters of a camera and transformation matrix,analysis the error via reprojection.Aiming at the problem of workpiece feature points extraction,The method of background subtraction is used to extract the edge line of the foreground area accurately,which effectively eliminate the influence of the complex background.Morphological processing is done in the process of establishing the foreground region ROI,which avoiding the influence of the morphological operation on the pixel coordinates of edges and feature points.We take the intersection point of workpiece' contour line as the feature point in workpiece rather than extracting the contour corner point from the picture of workpiece directly,that makes it is accurate and effective.Finally,we estimate candidate corresponding points by the constraint condition of epipolar line,and eliminate the mismatching points according to the similarity condition of the gray value.This means is quickly and accurately to find the best matching point.The experimental results show that the accuracy of positioning recognition can be controlled within 0.52469 mm when the camera is 1m away from the workpiece,which can meet the precision requirement of robot positioning.