Pose Estimation Of Target Based On Keypoints And Manipulator Grasping Control

In recent years,the rapid development of artificial intelligence and robot industry,it is an inevitable trend to use the manipulator to replace people to engage in some heavy and repetitive work.With the more and more complex working environment of the manipulator,the pose of the target in the practical application scene is complex and changeable,which brings a great challenge to the accurate grasping of the manipulator.Therefore,it is of great significance to study the position and pose estimation and grasping control of the target.In this thesis,the target's pose estimation based on feature matching and the grasping control method of manipulator are studied.In the aspect of pose estimation,feature matching method is introduced to recognize the target.Because the scale invariant feature transform algorithm and the speed up robust features algorithm exist the problems of less matching points and large time consuming,the grid-based motion statistics algorithm with high robustness and fast matching speed is selected for feature matching.Aiming at the problem of inaccurate matching in grid-based motion statistics algorithm,an improved grid-based motion statistics algorithm based on affine invariance principle is proposed.Firstly,the algorithm combines the robust random sample consensus algorithm to get a good initial base,solves the coefficients of all feature points relative to the initial base,and removes the false matching points by comparing the Euclidean distance among the coefficients.Secondly,the matching points are used as the index and combined with the depth image to transform the matching point set from the two-dimensional pixel coordinate set to the threedimensional camera coordinate set.Finally,the least square method combined with the matrix singular value decomposition algorithm is used to obtain the rotation translation matrix to estimate the pose of the target.In the aspect of accurate grasping control,a method based on projection angle is proposed.The method is mainly divided into two parts:(1)By constructing a virtual coordinate system,the target pose coordinate system is obtained at its projection angle to control the rotation pose of the end effector;(2)Through coordinate transformation,the position of the target center point is converted to the base coordinate system of the manipulator,and the grasping position of the end effector is controlled.Based on the visual manipulator system,the target recognition and grasping are verified by experiments.Firstly,the improved grid-based motion statistics algorithm combined with the least square singular value decomposition algorithm is used to estimate the pose of the book with feature points;Secondly,by calculating the projection angle of the target pose coordinate system in the virtual coordinate system,the end effector of the manipulator is controlled to move the specified pose and grasp the book.The experimental results show that the vision manipulator system can accurately recognize and grasp the book.