单目多视角全景视觉感知三维重构技术研究  

Research of 3D Reconstruction Based on Monocular Multi-view Panoramic Visual Perception

作　　者：郭伟青[1,2] 吴小刚[1] 汤一平[2] GUO Wei-qing;WU Xiao-gang;TANG Yi-ping(Zhijiang College,Zhejiang University of Technology,Shaoxing 312030,China;Information Engineering College,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学之江学院,浙江绍兴312030 [2]浙江工业大学信息工程学院,杭州310023

出　　处：《小型微型计算机系统》2019年第7期1525-1531,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61070134)资助

摘　　要：从图像中恢复出几何实体的三维形状是计算机视觉研究领域的一个核心课题.本文提出了一种基于单目全景成像的三维重构方法.通过单目多视角全景视觉传感器在一个摄像平面上同时获取从多个不同视角拍摄的被测物体图像,采用Otsu算法对图像进行分割,提取前景信息,用轮廓体素极坐标遍历得到几何实体的轮廓采样点数据,重构出几何实体的三维形状.单目多视角全景视觉传感器避免了通用多视角成像中摄像机的颜色系统以及内外参数难以保持一致的问题,减少了硬件成本和立体匹配的复杂度,增加了实时性;通过轮廓体素极坐标遍历得到物体的三维点云数据是一种明确、不含二义性的信息,具有较好的鲁棒性.通过实际系统应用验证,该方法是一种行之有效的三维重构方法.To recover 3D geometry from images is a core research in computer vision and computer graphics.An algorithm of Objectcentered 3D reconstruction based monocular multi-view images is presented.Using monocular multi-view sensor to gain one panorama picture with multi-view images of geometric object,segmenting this picture by Otsu algorithm,bringing up foreground information,acquiring outline sample points of the object by using traversal method of outline voxels based on polar coordinates,then reconstruct 3D geometric modeling.Using monocular multi-view stereo sensor can overcome the problem of usual multi-view images that the inner and outer parameters and color system of different cameras is difficult to keep the same,can reduce complexity of multi-view matching calculation and cost of hardware,can increase real-time performance.3D points of object reconstruction by outline voxel traversal of polar coordinates is clear and unambiguous,so it has good robustness.This method has been verified in the actual system.

关 键 词：以物为中心 单目多视角 三维重构 轮廓体素法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]