Research On Rapid Identification And Localization Of Metal Workpiece Based On 3D Vision

With the application of robots in more and more fields and the vigorous development of machine vision technology,more and more automatic and intelligent tasks can be completed by robots.In the face of the bin-picking requirements of small batches and multiple types of parts,making specific fixtures or production lines for the parts will make the production cost of the factory higher,so the research on vision-based intelligent identification and positioning technology of the workpiece becomes a necessary trend.In this paper,RGB-D camera is used as the source of visual information,aiming at the identification and positioning of automotive hardware fittings,object identification based on template matching,position and pose estimation based on point cloud registration,and hand-eye calibration based on nonlinear optimization and point cloud are studied.This paper mainly completes the following aspects:(1)Aiming at the problem of car hardware fitting recognition,an improved object recognition algorithm based on LINEMOD template matching is studied.In order to improve the recognition rate of LINEMOD template matching algorithm on reflective and texturless objects,this paper proposes to use gradient information and normal information of depth image as template features for template matching.At the same time,in order to improve the collection and production efficiency of the template,this paper analyzes the basic principle of ray tracing,and used Open GL to write an algorithm based on model rendering of multiple views,which can realize the rendering of the depth map of multiple perspectives of the workpiece and its corresponding point cloud.Experiments show that the improved method improves the recognition rate of the original method.(2)To solve the problem of 3D fast and accurate location of workpiece,the CUDA-based algorithm of KD tree ICP two-stage point cloud registration was studied.In order to obtain a better initial pose for accurate registration,the point centralization of two point clouds to be matched is firstly calculated,and a translation vector is formed through the two center points.The translation vector is used to zoom in the template point cloud and the measured point cloud,so as to achieve the coarse alignment of the point cloud.In the precise alignment stage,aiming at the time-consuming problem of ICP algorithm,a CUDA-based K-D Tree ICP algorithm was designed,which successfully improved the algorithm speed.The results of the experiment show that the proposed method can control the precision of accurate pose estimation within 2mm and shorten the time to about 30 ms,which is about 10 times faster than the original method,and can realize real-time 3d pose estimation of workpiece.(3)The hand-eye calibration algorithm of robot based on 3D vision is studied.In order to confirm the feasibility of the hand-eye calibration method based on the point cloud data with noise,the accuracy of calibration matrix estimation for solving the hand-eye calibration equation with linear calculation and nonlinear optimization under different noise conditions is compared.The simulated data experiment shows that the L-M algorithm can effectively solve the hand-eye calibration equation under the condition of high noise.Designed a kind of nonlinear optimization based on point cloud of hand-eye calibration process,using the artifacts that to be detected as the calibration objects.Rapid and convenient hand-eye calibration and calibration tool coordinate system were designed,the experimental results show that the proposed method of calibration error is under 3 mm,the visual system can satisfy the demands of metal workpiece robot 3D bin-picking.(4)Finally,a robot picking up system integrating the method proposed in this paper is designed and built,and a series of experiments are designed to verify the feasibility of the metal workpiece recognition and pose estimation system based on 3D vision proposed in this paper.