鱼眼变形立体图像恢复稠密深度图的方法.pdfVIP

第 23 卷 第 12 期 计 算 机 学 报 VoI. 23 No. 12 2000 年 12 月 CHINESE J. COMPU TERS Dec. 2000 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 鱼眼变形立体图像恢复稠密深度图的方法 贾云得 吕宏静 刘万春 （北京理工大学计算机科学与工程系 北京 100081） 摘 要 使用普通镜头立体视觉无法实现近距离或大视场的立体感知，鱼眼镜头立体视觉可以解决这一问题，因 为鱼眼镜头的视场角可以达到 180 . 鱼眼镜头获取大视场的同时也引入了严重的图像变形 . 文中讨论了一种使用 三个鱼眼镜头摄像机构成的多基线立体视觉系统恢复近距离大视场稠密深度图的方法 . 为了高精度恢复稠密深度 图，作者采用图像局部规范化方法、鱼眼变形图像校正方法、基于仿射变换区域匹配的相似性极小化准则，来实现 鱼眼变形立体图像对应性求解 . 文中最后给出了用真实物体和场景的大变形图像来恢复稠密深度图的实验结果 . 关键词 鱼眼变形，立体视觉，区域匹配，深度图 中图法分类号： TP391 Fish-Eye Lens Camera Stereo Vision for Dense Depth Map Recovery JIA Yun-De LU Hong-Jing LIU Wan-Chun （ Department of Computer Science and Engineering，Bei ing Institute of Technology，Bei ing 100081） Abstract When the cIearance between the ob ect and the stereo cameras is minimaI，stereo fish-eye Ienses can provide a fuII stereo view of the ob ect where conventionaI stereo Ienses faiI to do so. This is because fish-eye Ienses can provide a fieId of view which is approximateIy 180 . Fish-eye Ienses are use- fuI where a Iarge fieId view is reguired. Though fish-eye Ienses provide for a Iarge fieId of view，they introduce significant distortion in the image. This paper presents an aIgorithm for dense depth map re- covery by using muti-baseIine stereo vision with fish-eye Iens. We empIoy the normaIizing IocaI image， fish-eye distortion correction and area-based matching for obtaining both precise and stabIe estimation of correspondence. ExperimentaI resuIts obtained on severaI reaI stereo images are presented. Keywords fish-eye Iens，stereo vision，area-based matching，