| 题名 | 全向摄像机标定与三维重建研究 |
| 作者 | 邓小明 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2008-01-19 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 吴福超 |
| 关键词 | 全向摄像机 摄像机标定 除式模型 全景图 三维重建 Omnidirectional camera camera calibration division model panorama 3D reconstruction |
| 其他题名 | Omnidirectional Camera Calibration and 3D reconstruction |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 由于全向摄像机具有较大的视场角, 它在三维重建中具有广泛的应用前景. 基于全向摄像机的三维重建, 通常需要摄像机标定, 对应点匹配和三维结构计算三个步骤. 由于全向摄像机具有严重的畸变, 透视摄像机的标定算法都不再适用, 文献中也还没有很成熟的算法, 因此研究全向摄像机标定具有重要的理论意义和巨大的实用价值. 本文围绕全向摄像机的标定开展了比较深入的研究, 所完成的主要工作有： 1. 基于一维标定物的反射折射摄像机标定算法. 一维标定物是指一组任意两点距离已知的共线点. 在透视摄像机标定中, 需要对一维标定物限制做一些特殊的运动. 本文研究表明: 通过一维标定物的多次一般运动就能够标定反射折射摄像机的全部参数. 此外, 本文方法还能够给出镜面参数的解析表示以及一维标定物相对于视球中心的位置. 2. 基于平面标定物的反射折射摄像机标定算法. 本文方法避免了现有方法中对部分可见的二次曲线进行拟合的问题, 具有标定精度高、简单易用的优点. 本文的方法还能用于标定一些鱼眼摄像机. 基于用本方法标定的鱼眼摄像机, 完成了鱼眼图像的自动全景图拼接和漫游系统. 3. 除式模型的理论分析以及基于除式模型的全向摄像机统一标定算法. 除式模型是已被广泛应用的径向畸变校正模型, 但目前还没有关于它的理论分析报道. 我们首次对除式模型进行了较为深入的分析, 得到三个理论结果: (1) 除式模型可通过对反射折射摄像机的径向畸变校正模型做Taylor级数展开得到. (2) 建立了除式模型中的畸变系数与反射折射摄像机内参数的代数关系. (3)导出了l阶除式模型用于反射折射图像校正的误差上界, 该上界可用于在标定前选择模型阶数. 基于这些理论结果, 提出了全向摄像机标定的统一算法, 该方法能够通过单幅图像标定反射折射摄像机的所有内参数, 或通过多幅图像能标定一般全向摄像机. 4. 透视摄像机和径向畸变摄像机的系统自标定算法. 对于该摄像机系统的标定问题, 已有的方法要求事先标定部分摄像机的畸变系数. 本文工作表明: 对于包含一个透视摄像机和三个以上抛物面反射折射摄像机的系统, 通过极几何关系可线性地标定反射折射摄像机, 透视摄像机也可通过Kruppa方程进行线性标定. 在标定过程中, 仅用到反射折射摄像机的主点先验知识, 不需要对透视摄像机的内参数做出假设. 本文方法还能自标定由其它类型的反射折射摄像机或一般全向摄像机与透视摄像机构成系统. 另外, 我们给出一种使用已标定透视摄像机标定径向畸变摄像机的线性算法. |
| 英文摘要 | Omnidirectional cameras have the prominent advantage of very large field of view, and they are widely used. Since the camera calibration is one of the indispensable steps to recover 3D information from 2D images, the thesis is focused on omnidirectional camera calibration. The main work is summarized as follows: 1. Catadioptric camera calibration with 1D objects. A 1D calibration object is meant a segment with several known-distance markers (or points). Generally speaking, calibration methods with 1D objects are more flexible than those with 2D or 3D objects. Under the pinhole camera model, the calibration is not possible with free-moving 1D objects. In this work, we propose a catadioptric camera calibration method using 1D objects with five or more known points. Our method is capable of calibrating a camera when either the camera or the calibration object undertakes three or more general motions. In addition, the analytical solutions of the mirror parameter and the 1D object’s pose are also provided. 2. An easy calibration for catadioptric camera. We propose an easy calibration method for central catadioptric cameras with a 2D calibration pattern. The proposed method does not need any fitting of partial visible conic, and our method is easy and robust. 3. Radial division model revisited. Division model is a widely used imaging model for omnidirectional camera. However, no theoretical analysis about division model has been reported yet. We make a systematic study on division model, based on which a unified calibration method for both central catadioptric cameras and generic omnidirectional cameras is proposed. The main contributions are as follows: the division model is derived by Taylor expansion, and it is found that the catadioptric parameters can be computed with the division model. Then an upper bound of rectification error is derived for central catadioptric cameras, which could be used for the order selection of division model. The calibration method is shown to be able to calibrate all the catadioptric parameters from merely a single image of a 2D pattern. 4. Self-calibration of mixing omnidirectional and perspective cameras. Former calibration methods on mixing omnidirectional and perspective cameras assumed either the distortions or the intrinsic parameters of omnidirectional cameras were precalibrated. In this method, the radial distortion and the fundamental matrix can be simultaneously estimated with linear computations. With at least three paracatadioptric cameras, the perspective camera can be linearly calibrated by the Kruppa equations. The advantage of our method is that it can not only calibrate the central catadioptric cameras but also simplify a traditional nonlinear self-calibration of perspective cameras to a linear process. Moreover, a linear calibration method for an omnidirectional camera with a calibrated perspective camera is also derived. |
| 语种 | 中文 |
| 其他标识符 | 200418014628015 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6045]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 邓小明. 全向摄像机标定与三维重建研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2008. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论