Generalising the ideal pinhole model to multi-pupil imaging for depth recovery
Byrne, Brendan P. (2012) Generalising the ideal pinhole model to multi-pupil imaging for depth recovery. PhD thesis, Dublin City University.
Full text available as:
This thesis investigates the applicability of computer vision camera models in recovering depth information from images, and presents a novel camera model incorporating a modified pupil plane capable of performing this task accurately from a single image. Standard models, such
as the ideal pinhole, suffer a loss of depth information when projecting from the world to an image plane. Recovery of this data enables reconstruction of the original scene as well as object and 3D motion reconstruction. The major contributions of this thesis are the complete characterisation of the ideal pinhole model calibration and the development of a new multi-pupil imaging model which enables depth recovery. A comprehensive analysis of the calibration sensitivity of the ideal pinhole model is presented along with a novel method of capturing calibration
images which avoid singularities in image space. Experimentation reveals a higher degree of accuracy using the new calibration images. A novel camera model employing multiple pupils is proposed which, in contrast to the ideal pinhole model, recovers scene depth. The accuracy of the multi-pupil model is demonstrated and validated through rigorous experimentation. An integral property of any camera model is the location of its pupil. To this end, the new model is expanded by generalising the location of the multi-pupil plane, thus enabling superior flexibility over traditional camera models which are confined to positioning
the pupil plane to negate particular aberrations in the lens. A key step in the development of the multi-pupil model is the treatment of optical aberrations in the imaging system. The unconstrained location and configuration of the pupil plane enables the determination of optical distortions in the multi-pupil imaging model. A calibration algorithm is proposed which corrects for the optical aberrations. This allows the multi-pupil model to be applied to a multitude of imaging systems regardless of the optical quality of the lens. Experimentation validates the multi-pupil model’s accuracy in accounting for the aberrations and estimating accurate depth information from a single image. Results for object reconstruction are presented establishing the capabilities of the proposed multi-pupil imaging model.
Archive Staff Only: edit this record