Please use this identifier to cite or link to this item:
|Title:||Depth distortion under calibration uncertainty|
|Authors:||Cheong, L.-F. |
Structure from motion
|Source:||Cheong, L.-F., Peh, C.-H. (2004-03). Depth distortion under calibration uncertainty. Computer Vision and Image Understanding 93 (3) : 221-244. ScholarBank@NUS Repository. https://doi.org/10.1016/j.cviu.2003.09.003|
|Abstract:||There have been relatively little works to shed light on the effects of errors in the intrinsic parameters on motion estimation and scene reconstruction. Given that the estimation of the extrinsic and intrinsic parameters apts to be imprecise, it is important to study the resulting distortion on the recovered structure. By making use of the iso-distortion framework, we explicitly characterize the geometry of the distorted space recovered from 3D motion with freely varying focal length. This characterization allows us: (1) to investigate the effectiveness of the visibility constraint in disambiguating errors in calibration parameters by studying the negative distortion regions and (2) to make explicit those ambiguous error situations under which the visibility constraint is not effective. An important finding is that under these ambiguous situations, the direction of heading can be accurately recovered and the structure recovered experienced a well-behaved distortion. The distortion is given by a relief transformation which preserves ordinal depth relations. Thus in the case where the change in focal length is not well estimated, structure information in the form of depth relief can be obtained. Experiments were presented to support the use of the visibility constraint in obtaining such partial motion and structure solutions. © 2003 Elsevier Inc. All rights reserved.|
|Source Title:||Computer Vision and Image Understanding|
|Appears in Collections:||Staff Publications|
Show full item record
Files in This Item:
There are no files associated with this item.
checked on Dec 7, 2017
WEB OF SCIENCETM
checked on Nov 22, 2017
checked on Dec 10, 2017
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.