Research On Pose Recognition And Visual Servoing Technologies For Low Texture Parts

Intelligent manufacturing is the inevitable trend of future in industrial development.The intellectualization of industrial robots is the core problem of intelligent manufacturing.Compare with traditional control method of robots,visual servoing control technology which is a feedback control method of robot motion using visual information can provide higher flexibility,task accuracy and intelligent level.The existing visual servoing control is based on image feature points,which is difficult to be applied to the low-texture targets.Based on the existing research,a further study on the visual servoing control of low-texture targets with specific geometric features is carried out.The detection technology,pose recognition technology and corresponding visual servoing technology of parts with circular features and general planar features are studied.A robot visual servoing control platform is developed by combining the above technologies.Precise positioning and grasping of low-texture targets with geometric features are realized on the platform.The main contents of the paper are as follows:(1)Aiming at the problem that the point-feature-based visual servoing is difficult to apply in the grasping task of low-texture parts,a visual servo model of low-texture parts is proposed.The model takes the geometric features of low-texture parts as the basic control feature of visual servoing.By detecting,tracking and recognizing geometric features,the real-time state of features is provided for visual servoing control.Different visual servoing control methods are used to capture the geometric feature information.(2)Aiming to solve the detection problem of low texture parts,a method of detecting ellipse features based on gradient clustering is proposed,which guarantees the spatial continuity of the ellipses through the continuity of image edges.The plane feature measurement method of prior perspective invariant level set is proposed.By adding perspective transformation parameters into the solution process of energy functional and calculating gradient descent direction alternately,the minimum value of energy functional is obtained.The perspective invariance between current feature and reference feature is guaranteeded.Finally,the feasibility and superiority of the method are verified through experiments.(3)Aiming to solve the pose recognition problem of geometric features of low-texture objects,a method for identifying the pose of circular features based on intersection of cone surface is proposed.By analyzing the relationship between the camera model and the features of spatial circle,the analytic geometric equation of cone surface and sphere was established,and the normal vector of circular features in space is obtained by factorization.A planar pose recognition method is proposed based on frequency domain correlation analysis.The result of the affine transform is treated as the initial value of the set of fuctions,which are deduced on the basis of image moment.The homography matrix is obtained using an iterative solving algorithm.The pose is obtained through the singular value decomposition from the homography.(4)A position visual servo method based on the virtual axis positioning of circular features is proposed to solve the visual servo control problem of geometric features of low-texture objects.By defining the virtual axis on the plane of circular features,the six degrees of freedom of circular features are determined.A hybrid moment-based visual servo control method is proposed to overcome the problems in tranditional moment-based visual servoing method.The rotational degrees of freedom of the three high order moments are replaced with rotation vector to recieve the decoupling relationship between image features and motion variables.This helps the control scheme to get smoother trajectory,and eliminate the unstable state in the control process.(5)An integral visual servo control system based on contour and image moment is proposed to validate the effectiveness of the proposed methods.The correctness and feasibility of the proposed methods are verified through some experiments.