QQ客服 官方微博 反馈留言

题名	计算机视觉中摄像机定标和自定标研究
作者	邱茂林
学位类别	工学博士
答辩日期	1996-04-01
授予单位	中国科学院自动化研究所
授予地点	中国科学院自动化研究所
导师	马颂德
其他题名	CAMERA CALIBRATION AND SELF-CALIBRATION IN COMPUTER VISION
学位专业	模式识别与智能系统
中文摘要	摄像机定标的任务是，在某一个既定的摄像机模型下，经过对数字图 像进行处理，利用一系列数学变换和计算方法，求取摄像机模型的内部参 数和外部参数。通常摄像机定标所假设的摄像机模型是针孔模型。这不仅是 因为针孔模型简单，更主要的是视觉计算中的一些物理约束，譬如共面约 束和外极线约束，可以由针孔模型得到。针孔模型的简单性决定了其只能 在一定程度上表达实际的摄像机成像过程。于是就要考虑摄像机实际成像过 程的非线性畸变。实际的成像过程经过非线性畸变补偿后，才能更合理地 看作针孔模型成像过程。利用校正后的模型进行三维重建才能得到更高的精 度。因此，当对一个CCD摄像机进行定标时，除了确定针孔模型的参数 外，还要确定一个畸变校正模型。计算机视觉中的摄像机定标就是要利用 图像信息确定这两个模型。 现有的摄像机定标技术大体可以归结为两类：传统的摄像机定标方法 和摄像机自定标方法。在第一章中我们比较全面地回顾了现有的摄像机定标 技术。 传统的摄像机定标方法是利用一个标准参照物与其图像的对应约束关系 来确定摄像机模型的参数。为了提高计算精度，同时还需确定非线性畸变 校正模型的参数。在第二章中，我们首先分析了CCD摄像机成像过程中非 线性镜头畸变以及可能存在的其它非线性畸变因素。将人工神经网络中自组 织的思想应用于摄像机定标，提出了在摄像机定标时建立非线性畸变校正非 参数模型的方法。此方法与摄像机定标的参数化方法不同， 非线性畸变校 正是利用人工神经网络学习算法直接从输入训练数据得到，而不作任何数学 解析形式的假设，所得到的非参数畸变校正模型不仅考虑了CCD摄像机成 像过程中非线性镜头畸变因素，也考虑了摄像机成像过程中可能存在的其它 系统性的非线性畸变因素。分析与实验结果证实，该方法针对成像系统成 像过程中非线性畸变误差分析是一种可行的方法。利用这种非参数校正方法 对图像进行投射和对三维空间点进行重建的实验结果，表明了该方法的有效 性。 与传统的摄像机定标算法相比，摄像机自定标算法，无需知道三维控 制点的准确位置，只利用图像序列中的信息和由此得到的约束来求解有关摄 像机参数，摆脱了定标算法对标准参照物的依赖，使得实时地、在线地确 定摄像机参数成为可能。然而遗憾的是，已有的基于点对应的一些自定标 算法，其结果是不稳定的，更准确的说，在图像上含有噪音的情况下，自 定标的解受图像上噪音的影响非
英文摘要	Camera calibration, the art of determining intrinsic and extrinsic parameters of a CCD camera based on the information in digital images and therefore providing the link between metric and nonmetric cameras, is considered as a very important issue and a basic step toward stereo vision in computational computer vision theory. From a general point of view, the existing camera calibration techniques can be classified into two categories: traditional camera calibration and self-calibration. Chapter 1 gives a comprehensive review of these techniques. In traditional camera calibration, the task is to determine the parameters of a mathematical camera model, and furthermore the parameters of a distortion model if the nonlinear factors in the imaging process are considered such that better results can be achieved. So much attention has been paid to the techniques of determining more efficiently and reasonably the nonlinear distortion model. One aspect of our research work is also about the compensation of nonlinear distortions in the imaging process of a CCD camera. In Chapter 2, we will discuss several nonlinear distortion sources in the imaging process of a CCD camera and propose the nonparametric approach for camera calibration. Compared with parametric approaches for camera calibration, distortion surfaces are derived directly from the training data. Because of no any assumption of the analytical form of distortion surfaces, not only the nonlinear lens distortion but also some other nonlinear imaging errors, or the nonlinear systematic imaging errors are considered in nonparametric camera calibration. This gives a new approach to analyze nonlinear imaging errors of a particular imaging system. Experimental results both in image projection and 3D reconstruction confirm the effectiveness of the nonparametric approach. Self-camera calibration differs from traditional camera calibration in that a standard device often called "Setup" is indispensable in traditional camera calibration for obtaining the accurate 3D metrics while in self-calibration the camera parameters are tried to be computed just from constraints of image sequences without considering the 3D metrics. But what confrustrated researchers in self-calibration is that the solution is neither unique nor stable in front of noises. Therefore the main concem of reseachers in this area is to improve the robustness of the solution in front of noises in real images. Proposed in Chapter 3 are self-calibration methods based on more global features, for example, corresponding line segments and planes instead of corresponding points to improve the robustness of the solution of point-based self-calibration algorithms. It is impossible for us to get any constraint on two arbitrary corresponding lines each in one of two succesive images. But when line segments are divided into different groups of parallel ones, the constraint on corresponding lines can be derived based on th
语种	中文
其他标识符	367
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/5659]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	邱茂林. 计算机视觉中摄像机定标和自定标研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1996.

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。