Camera Self-calibration Using Multiple Geometric Constraints in a Single Image
WANG Meizhen,LIU Xuejun,LU Yue,LIU Dan
DOI: https://doi.org/10.3724/sp.j.1047.2012.00644
2012-01-01
Geo-information Science
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.