Integrating virtual objects into real environments presents a number of technical and aesthetic challenges which limit the impressiveness of current augmented reality applications. Although some of these challenges have recently been addressed, others are still active areas of research. The quality of the models and textures used, accurate geometric world alignment and the quality of illumination correspondence all need to be considered. Literature shows the latter to be the least mature field and is the focus of this paper. This project introduces a new photometric registration technique that makes use of image interest points. The technique attains illumination correspondence without the need for pre-calibration or unnatural calibration objects, instead natural image features such as shadow and object interest points are use. The operational complexity of the technique is low and real-time processing of live data is achieved.
|A multi-view interest point based approach to photometric realism within augmented reality systems.|