3D Panoramic Imaging for Virtual Environment Construction

The project is concerned with the development of algorithms for the creation of photo-realistic 3D virtual environments, overcoming problems in mosaicing, colour and lighting changes, correspondence search speed and correspondence errors due to lack of surface texture. A number of related new algorithms have been investigated for image stitching, content based colour correction and efficient 3D surface reconstruction. All of the investigations were undertaken by using multiple views from normal digital cameras, web cameras and a ”one-shot” panoramic system. In the process of 3D reconstruction a new interest points based mosaicing method, a new interest points based colour correction method, a new hybrid feature and area based correspondence constraint and a new structured light based 3D reconstruction method have been investigated. The major contributions and results can be summarised as follows: • A new interest point based image stitching method has been proposed and investigated. The robustness of interest points has been tested and evaluated. Interest points have been proved robust to changes in lighting, viewpoint, rotation and scale. • A new interest point based method for colour correction has been proposed and investigated. The results of linear and linear plus affine colour transforms have proved more accurate than traditional diagonal transforms in accurately matching colours in panoramic images. • A new structured light based method for correspondence point based 3D reconstruction has been proposed and investigated. The method has been proved to increase the accuracy of the correspondence search for areas with low texture. Correspondence speed has also been increased with a new hybrid feature and area based correspondence search constraint. • Based on the investigation, a software framework has been developed for image based 3D virtual environment construction. The GUI includes abilities for importing images, colour correction, mosaicing, 3D surface reconstruction, texture recovery and visualisation. • 11 research papers have been published