基于宽基线立体视觉的远距离三维重建  

Long-range 3D reconstruction based on wide-baseline stereo vision

作　　者：王炜强[1] 徐进[1] 杜歆[1] 刘济林[1] 

机构地区：[1]浙江大学信息与通信工程研究所,浙江省综合信息网技术重点实验室,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2010年第6期1073-1078,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目NSFC:60534070)

摘　　要：针对月球车视觉导航中远距离场景的三维重建问题,提出一种基于相机自标定的宽基线立体视觉方法.该方法首先提取立体图对的尺寸不变性(SIFT)特征并进行匹配,得到特征点对应关系;再使用外极线几何约束描述立体图像校正的直射变换,对其Sampson误差使用Levenberg-Marquardt(LM)算法求得最小二乘最优解,完成相机参数估计和图像校正;对校正后的图像使用种子像素视差扩张算法进行宽基线立体匹配,根据匹配得到的视差完成目标场景三维重建.实验表明:该方法解决了月球车相机的现场标定难题,并能够解决宽基线立体视觉面临的光照差异、透视畸变、遮挡等问题,对远距离山脉等自然场景的三维重建效果良好,重建精度较高.A self-calibration based wide-baseline stereo vision technique was proposed for remote scene＇s 3D reconstruction in vision based navigation of lunar rover. Correspondences between stereo rig were obtained by extracting and matching of scale-invariant feature transform （SIFT） features. Then with the epipolar geometrical description of rectifying collineation,the least-squares solution to Sampson error was achieved by Levenberg-Marquardt（LM） algorithm,which obtained the camera parameter and the rectified stereo rig. Following,the wide-baseline dense matching was completed using seeds＇ disparity expansion algorithm. Finally,target＇s 3D reconstruction was achieved on the calculated disparity. This technique is demonstrated to have good performance on a number of remote natural mountain scenes and solve the problems of field calibration,change in illumination,perspective deformations and occlusions in wide-baseline stereo vision.

关 键 词：宽基线 自标定 立体视觉 三维重建 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]