Abstract
The traditional approach of generating novel virtual views of an object or a scene is to render the appropriate 3-D model. The alternative is to render directly from the original images; this approach, which is based on pixel interpolation or reprojection, is called image-based rendering. In this paper, we describe a technique that enables virtual navigation within a complex environment using an image-based rendering technique. In particular, we make use of clusters of cylindrical panoramic images. Each cluster of panoramic images allows the user to smoothly navigate within a particular area, say within a single room. Having a collection of such interconnected clusters would enable the user to seamlessly navigate within a complex environment, such as an entire floor of a building, with each cluster representing a room. To achieve this goal, we examine a few techniques for image-based rendering using a cluster of cylindrical panoramic images to synthesize views from virtual viewpoints. We also describe our method for enabling smooth transition between clusters.
