Monocular Vision Positioning Method For Industrial Robot Assembly

The combination of camera and industrial robot enables industrial robot to get more intuitive access to environment,and the development of machine vision expand application scope of industrial robot.Currently,the application of industrial robot in assembly still stays in teaching model,which limits the pose of target workpieces.Usually,this method is adopted in mass production model to increases production efficiency.With the development of Internet of Things,flexible production line of small batch and customization will be future development trend,industrial robot without additional sensing will be unqualified.This article uses monocular cameras and industrial robots as research objects,explored in the area of camera and robot coordinate transformation,workpiece corner recognition and camera pose estimation.This paper demonstrated the application of monocular vision positioning method in robot assembly experiments,there is no artificial cooperation points in target workpiece,which verifies the feasibility of monocular vision positioning.In this paper,the kinematic model of industrial robot and camera model were established,Firstly.The kinematic model of SR10 C industrial robot is established by using homogeneous transformation method and D-H parameter.The kinematic positive solution of SR10 C industrial robot is deduced.Then,the perspective projection model of camera is established,and calibration principle and distortion principle are introduced.Camera is calibrated with the method of Zhang,and the intrinsic parameter is found.Hand-to-eye model is established based on robot and camera model,the transformation matrix is calculated by hand-eye calibration algorithm,the relationship between states and calibration error is also studied.Secondly,the method of corner detect for typical workpiece is designed.A workpiece with ridges is the target of assembly in this paper,and corners can be detected by ridges.First step,the mean filtering algorithm and Gaussian filtering algorithm are combined to filter out image noise,high quality image is got.Second step,Canny edge detection algorithm is used in image to get contour of workpiece,the ridge lines of workpiece in image is obtained by Hough line detection algorithm.Third step,the corners of workpiece can be obtained by compute cross line'intersections.Workpiece's corner can be detected precisely by this method.Thirdly,a new camera pose estimation algorithm is proposed based on the SoftAssign algorithm.This algorithm combines the advantage of orthogonal iteration and SoftAssign algorithm,it uses the object space collinear error of three-dimensional feature points and two-dimensional feature points as the gist of assigning feature correspondence and the optimization goal of calculating camera pose,to get the right correspondence between feature points and camera pose.This algorithm can make it not only in the case when the correspondences of 3D/2D feature points are one to one,but also in the case when there are both occlusion and clutter in one case.Results of the synthetic image and real image verify the algorithm's property.Finally,robot positioning experiment verified the proposed algorithm in this paper.On the experimental platform,the positioning process of industrial robot based on monocular vision is introduced.The camera accurately obtains the pose of multiple workpieces and controls the movement of the robot.Compared with other methods,the method proposed in this paper achieves the same positioning accuracy and positioning time.The source of the error is analyzed.