| Robot positioning and grabbing technology is an important research field of industrial automation.At present,the three-dimensional positioning technology of objects in industry mainly relies on machine vision.Among them,the binocular vision system is used commonly,and it can obatain the three-dimensional position and orientation of objects.Binocular vision system can recognize the position of objects in three-dimensional space.However,the binocular vision system requires high lighting effect in the working environment.When the light is darker,it will affect the positioning accuracy.Considering that the price of the line laser(line structure light)is normally lower than the industrial camera,replacing one of the cameras with a line laser can not only improve the system's working environmental adaptability,but also reduce the cost of the system in this way.In this paper,a line laser and industrial camera are used to build a monocular vision laser scanning system to identify the three-dimensional position and orientation of the object,and to combine the robot for positioning and grabbing.Firstly,the acquisition of three-dimensional point cloud and the kinematics of UR10 robot are studied.This paper analyzes and studies the principle of laser triangulation and binocular vision,lays the foundation for obtaining three-dimensional point cloud data.This paper builds the mathematical modeling of robots.The mathematical model is established and the kinematics derivation of forward and inverse solutions is conducted by taking UR10 robot as an example.The hand-to-eye calibration of the robot is researched in this paper,which lays a foundation for the subsequent positioning experiment of the robot.Secondly,the three-dimensional point cloud plane's feature extraction algorithm is studied.Because of the long acquisition time of the planar in 3D laser scanning point cloud and the low accuracy of plane extraction,this paper improves the traditional RANSAC algorithm and proposes a method based on iteration and sample points judgment optimization,which can identify and extract all planar features of the object quickly and accurately.Experimental results show that this algorithm can find the planar of the scanned object effectively and accurately.Then,the three-dimensional point cloud edge's feature extraction algorithm is studied.Because of the problem that the 3D point cloud edge' extraction is not good and the extraction time is too long,this paper proposes a fast and accurate improved method based on Gaussian map clustering edge extraction algorithm.By optimizing the traditional estimation normals and agglometrative clusters,the main component analysis is used to estimate the normals,then the covariance matrix eigenvalues of the nearest neighbor domain points of each point are analyzed to detect the edge characteristics,and the edge extraction and comparison experiments of different point cloud objects are carried out.The experimental results show that the proposed method can extract the edge features of point clouds quickly and efficiently.Finally,the robot positioning and grabbing experiment is conducted and verified.A monocular vision laser scanning system is built to acquire the three-dimensional point cloud of the object,and the point cloud data is processed to obtain its three-dimensional pose,and then the robot is controlled to position and grab the object,and to complete the experiment verification.This paper obtains the three-dimensional point cloud data of the upper surface of the object by constructing the experimental platform.The relevant algorithms to process the point cloud of the object are analyzed and optimized,and the three-dimensional pose information of the object is obtained.Then the robot is combined to locate and capture the object's plane.Finally,this paper carries out the experiment with the prism,the experimental results show that the proposed method has low cost,higher precision and fast speed,and has the certain reference significance. |
PDF Full Text Request