Visibility-based Forest Walk-through Using Inertial Level of Detail Models

Modern simulation and training applications often require visualization of large scenes with many complex objects. However, this places prohibitive requirements on graphics hardware. For example, while methods for rendering near photo-realistic vegetation scenes have been described in the literature, they require minutes of computation. In this paper we present scene and level of detail management (LOD) techniques for achieving interactive frame rates for a forest walk-through. The framework selects the LOD based on object visibility, in addition to projected size. Moreover, visibility is calculated at run-time so we can support dynamic scene modification, such as interactively altered trees due to wind, explosions or fire. We also introduce an inertial level of detail model to minimize popping artifacts – rather than instantly switching the LODs, discrete LODs are smoothly blended over a number of frames. The described techniques can be readily adapted for scenes other than forests, providing support for simulations involving both static and dynamic synthetic environments, whether natural (e.g., dense vegetation) or man-made (e.g., urban landscapes).