Research On Multi-camera Calibration And 3D Reconstruction Algorithm Based On Circumferential Distribution

In recent years,with the continuous progress and breakthrough of computer technology,some technologies related to computers have also been developed,including computer vision.Researchers have applied computer vision to many fields,such as industry,medicine and so on.Compared with the traditional 3D model building method,such as laser scanner and modeling software,3D reconstruction based on the image has the advantages of low cost and simple operation,it has become one of the hot directions of computer vision.In this thesis,3D reconstruction based on multi-camera is adopted,which includes image acquisition,camera calibration,extraction and matching of feature points,and 3D point cloud generation.Among them,how to calibrate multi-camera system is always a key and difficult point,the result of which has a great influence on the reconstruction accuracy.For the calibration method of the multi-camera system,the researchers have carried out a heated discussion and extensive research.Although some achievements have been achieved,there are many problems in existing methods,including the calibration process,low calibration precision,and high calibration cost.In this case,a multi-camera calibration method based on the transparent glass calibration board is proposed.The calibration for the intrinsic parameters of the camera,which is based on the Zhang's calibration methods is obtained separately.For the extrinsic parameters of the multicamera system with circumferential distribution,the transparent glass calibration plate is used as the calibration object and the calibration method based on the refraction projection model is proposed in this thesis.Zhang's method is modified and the calibration accuracy is improved.In order to complete 3D reconstruction by using the calibrated multi-camera system,the strategy is employed,which is that each adjacent two cameras form a binocular vision subsystem to obtain the three-dimensional information of the target object surface,and then unify the three dimensional points obtained by each subsystem to the same coordinate system,thus the global point cloud of an object is obtained.For each subsystem,the feature of the target image collected by the camera is extracted based on the point feature,then the epipolar constraint and feature similarity is used to match the feature points,which improves the efficiency and accuracy of the matching.Then,the triangulation principle is used to obtain the 3D information of each object's surface combined the matching point and the result of calibration.Finally,based on the Delaunay triangulation and texture mapping algorithm,the global point cloud is processed to generate the 3D model of the target object,thus completing the target of 3D reconstruction based on the multi-camera.Experiments prove the correctness and robustness of the algorithm proposed in this thesis.