Research And Application Of High Accuracy Calibration Method In Multi-view 3D Reconstruction

3D reconstruction technology is an important branch in the field of computer vision.Today,it has broad application prospects in various fields such as virtual reality,film special effects,biomedicine,and automatic driving.All visual problems are inseparable from the step of acquiring images by cameras.In order to determine the correspondence between images and the three-dimensional world,the first thing to do is camera calibration.The accuracy of the calibration results also directly affects the actual results of the three-dimensional reconstruction.In addition,with the increasing use of visual technology,traditional monocular or binocular vision has certain limitations in certain specific scenarios.At this time,a multi-vision vision system consisting of multiple cameras can solve problems such as insufficient field of view and blocking.Based on the research background and the theoretical research of camera calibration and three-dimensional reconstruction,this paper aims to achieve high-precision calibration of multi-cameras in multi-view three-dimensional reconstruction applications.The main work is as follows:(1)A sub-pixel edge location algorithm for circular mark points is proposed.For the shape feature of the circular target,firstly define the neighboring points in the direction of the edge gradient,then perform the gray interpolation on the neighboring points,and finally perform curve fitting and filtering.Combined with the traditional parameter fitting method,the sub-pixel level of the circle edge can be accurately located.Finally,the validity and accuracy of the method are proved by the comparison of the fitting error and the accuracy of the camera calibration experiment.(2)A method of camera calibration in a PnP-based multi-vision vision system is proposed.Since PnP is a method for calculating the pose of a camera by using three-dimensional points and its corresponding image point,the three-dimensional point coordinates of partial space are obtained according to the principle of binocular vision reconstruction,and the continuous reconstruction is performed in conjunction with the corresponding pixel coordinates on the subsequent camera image plane.The process of solving,merging and merging can gradually solve the rotation and translation matrix of any camera in the world coordinate system,and complete the calibration of the multi-camera.Finally,the distance measurement and other related experiments were performed by using the calibration results,and the accuracy was compared with the actual distance.The accuracy and feasibility of the method were proved.(3)A multi-view 3D reconstruction method based on incremental reconstruction is proposed.The process of multi-camera calibration and three-dimensional reconstruction in this paper is actually complementary.In solving the pose of the subsequent camera,it is necessary to use the three-dimensional point coordinates that have been reconstructed in the previous section.According to this principle,after obtaining the camera's extrinsic parameters at each step of calibration,the newly calculated point cloud information is fused with the previously known 3D points,and finally a multi-view 3D reconstruction result is obtained.At the same time,a BA optimization model based on the reprojection error function is constructed to optimize the reconstruction results.The reconstruction results obtained in the final experiment are basically in line with expectations.The comparison shows that the multi-vision vision system can obtain more abundant point cloud information,and the optimized results can also achieve high accuracy.