Optimum Alignment of Panoramic Images for Stereoscopic Navigation in Image-Based Telepresence System

Abstract

The addition of stereoscopic navigation to an image-based virtual environment is a desirable enhancement. This can be implemented by sampling the scene with a number of stereoscopic panoramas. In this regard, clusters of panoramas in a known spatial arrangement can be used to render omnistereoscopic views. However, slight misalignment between panoramas introduced by single-shot panoramic cameras must be corrected in order to capture the depth of the scene consistently in every direction. In this regard, a novel alignment correction method is proposed herein based on the dense disparity map between panoramas. This technique was successfully tested in the rendering of omnistereoscopic images in different scenarios and it is applicable to any off-the-shelf panoramic cameras. Unlike other panorama alignment methods, this is a featureless and uncalibrated solution to the alignment of closely taken panoramic snapshots. Furthermore, the omnistereoscopic rendering and aligning techniques proposed herein are computationally inexpensive alternatives for creating stereoscopic image-based virtual environments.