3d Reconstruction Aided Grasping System Based On Points And Circles Features

While upgrading our country's manufacturing,the demand of workpiece moving's intellectualized reconstruction becomes more urgent,especially for small and medium size manufacturing enterprises.At present,there are two major reconstruction schemes: first,combine the customized workpiece locating device and artificial calibration,which takes less equipment cost but more labor cost;second,use conveyor belt and articulated robot arm,which is more intelligent but expensive.In this work,we start from the pipeline workpiece's moving problem,and introduce a simultaneous identification of points and circles algorithm,to get low cost approach and less human intervention.Industrial environments generally have low texture and repeated details,but how to construct stable 3d models in low-texture secenes is still a hard problem in the field of machine vision.We represent the 3D target scene with sparse point cloud from multiple observations and extract accurate 3D circles to find the exact location of the workpieces.The introduced algorithm assists the industrial robot arm to grasp the rough pipeline workpieces.The major work of this thesis is as follows:1,Install a camera on a coordinate robot,and establish a real-time position feedback control system.Combined with the stable external parameter calibration method,the computing platform can obtain the world position of each frame,which gives the system the basis of monocular accurate3 d modeling.2,Feature-based modeling algorithm reconstructs the secenes from the sequence of images and gives a preliminary description.We explore the appropriate feature method with features' descriptors in the industrial environment,and optimize the feature matching process according to prior information.Bayesian filter is used to avoid the mismatching caused by the deceptive structure.Update spatial points with nonlinear optimization in image sequence,in which the phased optimization method can further filter the wrong matches and improve the accuracy of points' position.The fundamental points construction pipeline make the location of workpieces visible.3,Based on the point cloud,we explore the 3d circle modeling method.The pipe workpieces are represented by the 3d circles.We divide the generated point cloud into octree and set the condition to begin circles' initialization.The projection relations of 3d circles are analyzed to calculate the correspoinding 2d oblique elliptic equations in frames.Use projection and level set to find the 2d expression from a 3d circle,and then optimize the 3d circle's parameters.By the 3d circle modeling,all the pipeline workpieces' top surfaces are represented by the 3d circles,which can be grabbed and placed in order by the industrial robot arm.Verification experiments are performed on public data sets and collected target scene sequences,which prove that the proposed 3D reconstruction system based on space point and circle features,meets the modeling accuracy requirements and can aid grasping target workpieces.