基于单目视觉三维重建系统的设计与实现  

Design and Implementation of 3D Reconstruction System Based on Monocular Vision

作　　者：吕立 姚拓中[1] 宋加涛[1] 肖江剑[2] 王建军 L Li;YAO Tuozhong;SONG Jiatao;XIAO Jiangjian;WANG Jianjun(School of Electronic and Information Engineering,Ningbo University of Technology,Ningbo,Zhejiang 315211,China;Ningbo Institute of Industrial Technology,Chinese Academy of Sciences,Ningbo,Zhejiang 315201,China;Ningbo Jiangdong Quantuo Electronics Co.,Ltd.,Ningbo,Zhejiang 315040,China)

机构地区：[1]宁波工程学院电信学院,浙江宁波315211 [2]中国科学院宁波工业技术研究所,浙江宁波315201 [3]宁波江东泉拓电子有限公司,浙江宁波315040

出　　处：《计算机工程》2018年第12期233-239,共7页Computer Engineering

基　　金：国家科技支撑计划(2015BAF14B00);浙江省自然科学基金(LR13F020004;LQ15F020004);宁波市自然科学基金(2015A610132);宁波工程学院校级科研项目(2016032);浙江省公益技术应用研究计划(2016C33255)

摘　　要：视觉重建方法多数存在设备构成复杂、精度低、耗时高等问题。为此,提出一种单台手持式数码相机恢复场景的三维模型方法。采用经典的对极几何初始化相机位姿,使用改进Pn P算法计算多帧图像位姿,通过选取新图像鲁棒特征点扩建场景,并运用基于局部关联帧的光束平差法优化摄像机位姿和三维点位置。实验结果表明,与传统的光束平差法相比,该方法能够有效节约时间,具有较强的鲁棒性。Most of the visual reconstruction method have problems such as complicated equipment,low precision,and high time consuming.A three-dimensional model method for recovering a scene by a single handheld digital camera is proposed.The camera pose is initialized by the classic epipolar geometry.The improved PnP algorithm is used to calculate the pose in multi-frame images.By selecting new image robust feature points to expand the scene,the accuracy of the scene is improved,and the bundle adjustment method based on local correlation frames is used to optimize the camera pose and 3D point position.Experimental results show that compared with the traditional beam adjustment method,the method can save time effectively and has strong robustness.

关 键 词：单目视觉 三维重建 关键帧 摄像机位姿估计 局部光束平差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]