地球信息科学学报

地球信息科学学报 ›› 2012, Vol. 14 ›› Issue (5): 644-651.doi: 10.3724/SP.J.1047.2012.00644

• 遥感技术与应用 • 上一篇    下一篇

多几何特征约束的单幅图像相机自标定方法

王美珍, 刘学军*, 卢玥, 刘丹   

  1. 南京师范大学虚拟地理环境教育部重点实验室,南京 210046
  • 收稿日期:2012-04-10 修回日期:2012-09-17 出版日期:2012-10-25 发布日期:2012-10-25
  • 通讯作者: 刘学军(1965-),男,陕西合阳人,博士,教授,主要研究方向为DEM不确定性分析、影视GIS等。 E-mail: liuxuejun@njnu.edu.cn E-mail:liuxuejun@njnu.edu.cn
  • 作者简介:王美珍(1983-),女,江苏淮安人,博士,讲师,主要研究方向为影视GIS、单像解析等。E-mail: wangmeizhen@njnu.edu.cn
  • 基金资助:

    国家支撑计划项目(2012BAH35B02);江苏省高校自然科学重大基础研究项目(10KJA420025);江苏省测绘科研项目(JSCHKY201204)。

Camera Self-calibration Using Multiple Geometric Constraints in a Single Image

WANG Meizhen, LIU Xuejun*, LU Yue, LIU Dan   

  1. Key Laboratory of Virtual Geographic Environment, Ministry of Education, Nanjing Normal University, Nanjing 210046, China
  • Received:2012-04-10 Revised:2012-09-17 Online:2012-10-25 Published:2012-10-25

PDF (PC)

赞

被引次数

17 | 1

摘要：

目前,图像获取设备及方式呈现多样化趋势,获取的图像数量、重叠度等不完全具备传统摄影测量方法应用的需求。若要获取图像中包含的丰富场景几何信息,就需要发展依赖图像刻画内容的相机自标定方法,建立起图像与现实场景之间的桥梁。故此,本文提出了利用单幅图像中几何约束条件的相机自标定方法,并顾及多个几何特征约束,根据方差合理地为其设定不同的权重,以提高相机的标定精度。本文总结了现有现实场景中几何特征及可由其求解而得的相应不变量,以及几何特征对相机内参的约束关系。在对图像刻画场景中多个平面及几何特征进行编码的基础上,利用不同组合的几何约束条件求解相机内参,根据多次提取每一种几何特征组合求解内参的方差确定本次计算内参所占权重,综合每一组几何特征所确定的内参及其权重综合计算最终的相机内参值,从而实现多几何约束相机自标定方法。通过室内外场景验证测试,结果证明,本方法具有可用性、便捷性和较好的鲁棒性。

关键词: 多几何特征约束, 自标定, 单幅图像, 几何特征, 不变量

Abstract:

Camera self-calibration is a key step to acquisition 3D space information from 2D image, and it is always one of the important issues in photogrammetry. However, present methods for camera self-calibration need two or more images and/or their corresponding points. With the development of digital devices for image taken and (wireless) network, a method not depending on digital device, images taken process, or multiple images, is badly needed. Consequently this paper presented a novel method that makes full use of various geometric constraints to realize reliable camera calibration for a single image. Firstly, this paper summarized various geometric constraints and invariants for the existing camera self-calibration method. Secondly, in order to build the relationships among geometric constraints for calibration, we coded for different planes and geometric features in an image. Because variance represents the error distribution, it can be considered as the determinant. In this paper, we obtained the variance of different combination of geometric features for camera calibration by means of fitting each groups of geometric features for thirty times, and then depended on the variance above to determine the weight of each camera's internal parameters. Finally, based on each camera's internal parameters, here we only focus on foci length, and their corresponding weights, the ultimate results are computed. Two images which depict inside and outside scene respectively were chosen to test the usability of our methods. In order to avoid the influence of image distortion, we corrected it using amethod we proposed in another paper before tests. The test results show that: 1) the weighted method gave a more stable result, relative to the result of each group geometric constraints, that is one group's relative error is two high and in other may be lower; 2) the weighted method obtained a higher accuracy result than the mean of all groups. The results of verification testing for the two images of the indicated that our weighted method can comprehensive employs variety of geometric constraints in single image, in the other side, it also takes their corresponding variance into account. It makes full use of the variety, usability and stability of geometric constraints. It can be employed to images depict indoor and outdoor which contains more geometric constraints.

Key words: self-calibration, multiple geometric constraints, invariant, geometric constraints, single image