Research On Key Technologies And Design Of 3D Vision System Based On Line Structured Light

Three-dimensional vision technology increases the flexibility and intelligence of automated production.It also improves production efficiency and reduces labor intensity.In the industrial automation production process,the automatic feeding of the workpiece is an indispensable process.In this paper,the three-dimensional vision system has been taken as the research object.The calibration method of the threedimensional vision system and the identification and positioning method of scattered materials have been studied.By calibrating the three-dimensional vision system and processing the images acquired by the 3D vision sensor,the workpieces in the scattered materials could be identified,positioned and grasped.The paper has completed tasks.The calibration of 3D vision system has been studied.And a non-contact 3D vision system hand-eye calibration method has been proposed.Aiming at the feeding requirements of automotive seat frame welding and the characteristics of threedimensional line structured light measurement methods in industrial sites,a threedimensional vision system solution based on line structured light has been designed.Secondly,a 3D vision sensor model has been established.The 3D vision sensor calibration software has been designed to complete the calibration of the internal,external parameters and light plane parameters of the 3D vision sensor.Finally,the proposed non-contact 3D vision system hand-eye calibration method is used to complete the 3D system hand-eye relationship calibration.This method uses a cone as a calibration block,collects the image of the calibration block through a 3D vision sensor,processes and calculates the image based on a computer program written by Halcon.Obtaining the coordinates of the characteristic points of the calibration block in the 3D vision sensor coordinate system,and using this method to calibrate the threedimensional system improved the calibration efficiency and precision of the industrial site.The preprocessing method of 3D point cloud image has been studied.The preprocessing of scene point cloud image has been realized.Aiming at the image of scattered workpieces,the direct filtering method is used to filter out background and irrelevant point cloud data.The voxel grid mesh method is used to streamline the point cloud data.The Kd-tree algorithm is used to establish the point cloud topological relationship and implement the point cloud image.The K-neighborhood is quickly queried,and outliers are liminated using statistical methods.The above processing greatly reduces the point cloud data and improved the image quality.Based on the above work,the point cloud normal vector has been estimated using the plane principal component analysis method.The normals are adjusted for consistency to prepare for subsequent workpiece identification and positioning.The method of positioning and identifying scattered materials has been researched.The positioning and identification of scattered workpieces based on the 3D reconstruction template is realized.A fast Delaunay triangulation algorithm is used to reconstruct the workpiece template.And PPF feature descriptors are extracted based on constraints.A workpiece positioning and recognition algorithm based on Huff voting and ICP has been proposed.First,Huff voting is used to identify and coarsely locate the workpiece.Then the result of the rough positioning is used as the initial value of the ICP algorithm to precisely locate the workpiece.The positioning accuracy coefficient of the method in the workpiece positioning recognition experiment is 0.902,and the detection time is 4.43 s.The methed takes into account both accuracy and efficiency.The paper designed human-computer interaction software for disorderly grasping of materials and conducted online grasping experiments.A weighted evaluation method of workpiece grabbing based on the matching score and the average value of z-values has been proposed to provide a basis for the priority grabbing of workpieces.The experimental capture success rate is 98.7%,which meets the needs of the project.