Visualizing 3D scenes using non-linear projections and data mining of previous camera movements

Abstract

We describe techniques for exploring 3D scenes by combining non-linear projections with the interactive data mining of camera navigations from previous explorations. Our approach is motivated by two key observations: First, that there is a wealth of information in prior explorations of a scene that can assist in future presentations of the same scene. Second, current linear perspective camera models produce images that are too limited to adequately capture the complexity of many 3D scenes. The contributions of this paper are two-fold. First, we show how spatial and temporal subdivision schemes can be used to store camera navigation information that is data mined and clustered to be interactively applicable to a number of existing techniques. Second, we show how the movement of a traditional linear perspective camera is closely tied to non-linear projections that combine space and time. As a result, we present a coherent system where the navigation of a conventional camera is data mined to provide both the understandability of linear perspective and the flexibility of non-linear projection of a 3D scene in real-time. Our system's generality is illustrated by three visualization techniques built with a single data mining and projection infrastructure.