Design Of Binocular Vision Guidance System Based On Structured Light

With the rapid development of modern industrial manufacturing technology,the combination of visual inspection and robot technology is the main direction of intelligence industrial systems.Measurement based on machine vision has non-touch,high speed,high accuracy and other prominent advantages,especially in 3D measurement technology.However,there is no perfect solution in complex scenes such as stacked parts,so it is necessary to conduct in-depth research on this issue.In this paper,determining the detection method of object pose in the scene,especially in the scene of complex parts.Based on this,a binocular vision guidance system based on structured light is designed to solve the problem of sorting parts.The guidance system is constructed by the combination of binocular stereo vision technology and structural light technology.The system consists of three parts: double target,three-dimensional reconstruction and pose estimation.There are some difficulties in these links: calibration algorithm of binocular vision system,structured light projection scheme and feature matching technology,and specific implementation method of three-dimensional pose estimation.Aiming at the calibration algorithm of binocular vision system,analyzing the conversion relationship between coordinate systems according to the imaging principle of the camera,and the cause of the image distortion.Calibrating the binocular vision system by Zhang's plane calibration algorithm,and to perform distortion correction and extreme line correction on the obtained images.Aiming at the structured light projection scheme and feature matching technology,an improved projection scheme combining gray-code and phase-shift structure light is proposed.Decoding by projecting a six-bit inverse-positive gray-code pattern.To solve phase by threestep phase shift,a noise compensation algorithm based on the ? phase shift is proposed.Performing periodic misalignment correction on the phase.It realizes sub-pixel feature matching by the method of linear fit.Finally,the 3D point cloud reconstruction of the scene is realized based on the principle of triangulation.Aiming at the specific implementation method of 3D pose estimation,the Halcon-based model making and pose estimation algorithm is introduced to filter,segment and surface reconstruct the 3D point cloud of the scene.Finally,model making and pose estimation were achieved.Finally,a set of binocular vision guidance software and hardware system based on structured light is designed and developed.The hardware consists of industrial cameras,lenses,projectors,brackets and industrial computers.The software includes binocular calibration,model making and binocular vision guidance system interface.Quick calibration of the system with higher precision is achieved by using the calibration plate;The test of the bottles of complex curved surface successfully realized the model making and the estimation of multiple target poses of the model;The point cloud of the calibration plate plane reconstruction has reached the level of decimillimetre;The system uses a CPU-based multi-thread optimization method,which improves the efficiency of single-thread operation by 59.5% and runs continuously for 16000+ times.In summary,the experimental results verify the effectiveness,feasibility,accuracy,efficiency and stability of the system.