Image quality for driving simulation experiments
Roland Brémond, Gilles Gallée · 2007
Abstract: The databases which are currently used in driving simulators do not take lighting into account in a realistic way. First, the visualization of surfaces use photographic textures. These textures are modified by computer graphics designers in order to visually match with a specific lighting environment, without photometric or colorimetric control of any kind. For instance, a database build from daytime pictures will need contrast, brightness and color modifications for a night-time simulation with road lighting. Secondly, the textures are stored as pixel intensities, instead of physical units, which yelds display dependant rendering. Yet, color and contrast fidelity would make driving simulators useful for many new applications, specially for road safety, which involves visibility issues (driving through fog, for instance). This motivation led us to test if current lighting computation techniques could be used in real time applications. Realistic rendering of contrasts and colors is a fast evolving topic in the field of computer graphics. Physical modeling of light transport, using the radiosity method for instance, is a standard approach to get realistic rendering in synthetic images. Such computations would improve the realism of driving simulators. Unfortunately, the algorithms are far from compatible with real time. This is due to high computation times, but also to the geometric representation of the surfaces, which need to be meshed (like in finite element methods). It leads to amounts of polygons that cannot be handled by current rendering technologies. In this paper, we propose a methodology allowing the use of lighting calculation techniques as a pre-process of database rendering on driving simulators. Our goal is to allow contrasts and colors to quantitatively match real world values, so as to use driving simulation in visibility and lighting evaluation experiments, and to increase the use of driving simulation for road safety studies. The presented methodology was applied on a database of the Fourvieres tunnel (Lyon, France).