Research On Industrial Robot Workpiece Localization Method Based On Three-dimensional Vision

With the era of industry 4.0 coming and the rapid development of automation industry,the traditional produce mode is being gradually declining,the new industrial automation produce mode are growing very fast,and the assembly of industrial grabbing is an indispensable segment.Aiming at the demand of speed,efficiency,and accuracy in diverse mass-production mode,using automated grabbing based on industrial robot instead of traditional manual mode can improve the efficiency and accuracy.The essence of automated grabbing of industrial robot is the process of identifying,locating,and grabbing and positioning the workpiece to a specified location.In this thesis,the method of positioning of industrial robot has been studied based on 3D vision.The essence of positioning of industrial robot is through the 3D point cloud matching to achieve the purpose of positioning the workpiece.The essence of 3D point cloud matching is to select the 3D point cloud matching algorithm.This thesis takes industrial robot as the background,and takes the scattered stack workpiece as the research object,and based on the 3D image matching space makes an in-depth research on the 3D point cloud matching algorithm.It has a wide range of applications and important scientific significance.The main works of this thesis are as follows:(1)Create an RGBD data model of the workpiece.To aquire the RGBD data of scattered stack workpiece by 3D vision sensor Kinect,and extract the depth information based on RGBD data.And then aquire its point cloud of the scattered workpiece according to the depth information.The workpiece's point cloud background segmentation is based on its background,and eliminates the background that is not expected.And then aquire its 3D point cloud.(2)To extract the feature of scattered stack workpiece point cloud based on pair-wise point feature.According to the point of the workpiece point cloud and its normal vector information,this thesis presents a pair-wise point feature describing the 3D point cloud of the workpiece,and makes it better to describe the pair-wise point feature extraction method.And use the stored pair-wise point method of hash table data structure to realize the fast searching and accessing of pair-wise point features stored in the hash table,and improve the real-time performance of pair-wise point feature search.(3)Pair-wise point feature matching of scattered stack workpiece point cloud based on voting algorithm.Based on the research background of industrial robot workpiece grabbing,this thesis analyzes the classic 3D point cloud matching algorithm.Combined with the features of point cloud of the mentioned points above,this thesis proposes a strategy of pair-wise point feature matching based on voting algorithm.Using ubuntu 14.04 LTS 64-bit operating system and based on Qt Creator programming environment platform and combine with PCL(Point Cloud Library)to write programs,making positioning experiments of scattered stack workpiece.Firstly,makes the target workpiece to learn.Then makes pair-wise point feature matching and pose hypothesis voting.Finally,the poses are estimated by cluster pose hypothesis votes to achieve the purpose of scattered stack workpiece positioning.