Research On Multi-Workpiece Recognition And Grabbing Based On Geometric Feature Fitting And Decision Fusion

In recent years,it has been becoming a trend to integrate machine vision into robotics technology and apply it in production.At the moment,this technology has been widely used in assembly,sorting,packaging and defect detection.However,in those areas,the workpieces are placed on a fixed plane neatly.When the workpieces are stacked together,the traditional method is to use vibration screening mechanism and escapement mechanism to arrange the workpieces in order,and then use machine vision technology to guide the robot to complete its work,but this method will damage the surface of the workpiece,and the cost of the mechanical mechanism is high.It cannot meet the production rhythm demand in real time,and it is difficult to achieve flexible production.To solve the problem of grasping stacked workpieces,this paper studies a workpiece recognition and grasping technology based on binocular stereovision.Firstly,Using MATLAB toolbox for camera calibration to get internal and external parameters matrices.In this paper,the geometric feature is used to recognize the workpiece.First,the image is preprocessed by Gamma transform and bilateral filtering,then the Canny operator combined with edge connection is used as the edge extraction algorithm,which can extract the complete edge of the image.Finally,the elliptic feature fitting method based on mid-chord points is used to detect the workpiece.Secondly,support vector machine is used to classify the detected workpiece,and a classifier with variable weights is proposed,which combines LBP features with HOG features and has higher accuracy in workpiece recognition.An improved hand-eye calibration method is used in this paper,and a binocular ranging model is proposed.The workpiece is reconstructed three-dimensional based on geometric feature matching,then the position and posture of the workpiece are calculated.Finally,the overall scheme of binocular vision positioning system is designed according to the actual needs.Then,the hardware used in the scheme is designed and selected,and eye-to-hand system is selected.using VS2015 and MATLAB to complete the design of binocular vision positioning system software,and build an experimental platform.Through experiments to analyze the positioning accuracy and feasibility of the system,the grasp of stacked workpieces is completed,and the feasibility of the algorithm is verified.