双目立体视觉下破损黑陶三维重建方法仿真  

Simulation of 3D Reconstruction Method of Damaged Black Pottery under Binocular Stereo Vision

作　　者：潘春花[1] 赵德才 PAN Chun-hua;ZHAO De-cai(Computer College,Qinghai Minzu University,Xining Qinghai 810007,China;Physics and Electronic Information Engineering College of Qinghai Minzu University,Qinghai Xining 810007,China)

机构地区：[1]青海民族大学计算机学院,青海西宁810007 [2]青海民族大学物理与电子信息工程学院,青海西宁810007

出　　处：《计算机仿真》2024年第9期223-227,共5页Computer Simulation

摘　　要：在三维重建过程中,由于受环境光线变化、物体透明度、遮挡等因素影响,导致重建结果不准确或不完整,使得视觉效果较差。为了提高三维重建精度,提出一种基于双目立体视觉的破损黑陶三维重建方法。通过转换世界-相机-图像-像素坐标系,标定双目立体视觉相机内外参数,引入最小生成树方法立体匹配左右图像对,计算图像像素绝对强度、梯度、匹配截断值对黑陶重建曲面平滑度的影响,得到连续稠密的视差图,使用最小二乘曲线拟合法平滑处理黑陶三维点云数据,将游离在曲面外的点映射到曲面中,通过三角剖分实现破损黑陶三维表面重建。实验结果表明,所提方法能够有效提高重建后黑陶表面平滑度,且重建结果尺寸精准,拟合度高。In the process of 3D reconstruction,the change of ambient light,object transparency,occlusion and other factors may lead to inaccurate or incomplete reconstruction results.To improve the accuracy of 3D reconstruction,this paper presented a method of 3D reconstruction for damaged black pottery based on binocular stereo vision.By converting the world-camera-image-pixel coordinate system,we calibrated the internal and external parameters of the binocular stereo-vision camera,and introduced the minimum spanning tree method to match the left and right image pairs stereoscopically.Then,we calculated the impact of the absolute intensity,gradient and cut-off value of image pixels on the smoothness of the reconstructed surface of black pottery,thus obtaining a continuous dense disparity map.Moreover,we used the least square curve fitting method to smooth the 3D point cloud data of black pottery,thus mapping the points floating outside the surface onto the surface.Finally,we completed the 3D reconstruction of damaged black pottery through triangulation.Experimental results show that the proposed method can effectively improve the smoothness of the reconstructed surface of black pottery,and the reconstructed results have accurate dimensions and high fitting degree.

关 键 词：双目立体视觉 三维重建 坐标系转换 匹配代价聚合 立体匹配 三角剖分 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]