Camera self-calibration for augmented reality

Abstract

One of the characteristics of a good Augmented Reality (AR) system is to be able to register virtual objects correctly onto the specified location in the real world. A displacement from this specified location is called registration error. This error can be caused by several factors, such as system delay, optical distortion and poor camera calibration. The problem of poor camera calibration in AR has always been overcome by employing careful camera calibration steps with the use of a specific known object. However this task is time consuming and mostly performed offline. This dissertation aims to develop an AR system known as a SCAR (Self-Calibration for Augmented Reality) system, which incorporates a self-calibration of a camera so that the system updates the camera intrinsic parameters whenever they change. The SCAR system incorporates an algebraic approach to self-calibration where it can solve camera parameters based on only three views and requires only the fundamental matrices as the inputs. The solution proposed here can be used for any AR system that uses visual-based tracking. Several pre-calibration stages including feature detection, point correspondence matching, and fundamental matrix estimation are developed. The problem of inaccuracy with general corner detector has been identified and a new algorithm, which combines Harris and SUSAN corner detectors, has been suggested. This hybrid detector increases the corner detection accuracy and reduces localisation errors of up to several pixels. A novel point correspondence matching has been developed, which is based on motion vector and simple statistic calculation. The matching process is efficient and capable of removing outliers from corner detection as well as maintaining a good number of correct matches even in the event of occlusion. Lens distortion is a common problem in visual-based tracking. This problem is dealt with by solving general epipolar constraints in order to simultaneously solve for the distortion parameters and fundamental matrix using MAPSAC algorithm. Finally the SCAR system is compared with the ARToolKit calibration procedure and has proven to produce reliable results with better flexibility. The theoretical aspects of critical motion, which may exist between pairs of views, are also discussed. A new measure of the criticalness of the motion for the case of parallel camera axis followed by rotation along the principal axis is also presented.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata