3D Reconstruction Of Parallel Robot Working Environment Based On Multi-Vision And Its Application

Under the influence of the "Made in China 2025" strategic action plan,China is in a phase of high-quality development in the field of intelligent robotics.Some of the traditional robots are based on offline programming of reciprocal movements.As modern computer vision technology continues to upgrade,more and more tasks that can be performed through the human eye are being replaced by machine vision,providing ideas for vision-based 3D reconstruction and applications of modern robots.In this thesis,3D reconstruction of targets and robotic grasping based on multi-vision systems are investigated using parallel robots as application tools.Firstly,this thesis studies the camera imaging principle and the conversion relationship between different coordinate systems.Selects the main hardware equipment on the basis of the monocular and binocular vision models,and designs and assembles a multi-ocular stereo vision system in combination with the system software.The images captured by the three cameras were calibrated using the Zhang Zhengyou calibration method.To ensure the calibration accuracy of the multiocular cameras,manual calibration is completed using the MATLAB2018 b Tool Box toolbox,laying the foundation for the subsequent 3D reconstruction work.Then,three pre-processing methods,image greyscale,image filtering and histogram equalization,are applied to the initial image taken after the camera has been calibrated to obtain an image with a high signal-to-noise ratio.The RANSAC algorithm is used to improve the screening of false matching points according to the traditional SURF algorithm using the principle of parallax gradient to address the problem of easy false matching during feature matching.Through the feature matching comparison experiments of different algorithms,it is proved that the improved algorithm not only has higher matching accuracy,but also reduces the program running time.It has higher robustness and improves the target recognition ability.Secondly,according to the different quality and computing efficiency of parallax maps generated by different stereo matching algorithms.In this thesis,SGBM semi global stereo matching algorithm and BM string search algorithm are used for stereo matching of cylinder bottles under the same conditions.SGBM was selected as the targeting algorithm for this thesis by comparing the experimental data.The original image and the parallax image after filling the cavity were reconstructed by Visual Studio 2017 software to complete the 3D point cloud reconstruction,and the point cloud stitching was carried out by using the ICP principle to stitch the three groups of point cloud images from different viewpoints into a complete 3D point cloud image of the target object.Finally,this thesis designs a parallel robot operating platform based on a multi-vision system.Establish communication between computer and parallel robot according to TCP/IP address.The hand-eye calibration of the robot was performed using the eye-in-hand model,the relative positions of the camera and the end of the robot arm were determined.The research of the thesis was demonstrated through experiments that the parallel robot can accurately complete the positioning and grasping operation.