The main sources of illumination in outdoor scenes are the sun and sky and an essentialaspect of realistic outdoor scenes is correct rendering of the sky 4. Althoughthere have been many developments in the field, rendering an accurate representationof the sky at a specific location and time is still a challenge 8.Many researchers have used photography to derive empirical sky models.
Typically,the sky is captured through a single photograph taken with a calibrated camera.The photograph is then used to derive a customized mathematic model 8. Thismethod is quite tedious, camera dependant and provides only a limited dynamicrange of luminance values.
Another way to develop a sky model is to use measured data directly 4 and asubstantial amount of data is accumulated from long term measurements throughsky scanners in various parts of the world. The International Daylight MeasurementProgram (IDMP) was organised by the International Commission on Illumination(CIE) to collect worldwide information on daylight availability 4. These skylightmodels produce reasonably predictive skies for a variety of locations 9. Several researchers have simulated atmospheric effects. For instance, Klassen used aplanar layer atmospheric model and single scattering to simulate sky colour 10. Asimilar simulation that uses a spherical atmosphere with air density changing exponentiallywith altitude was employed by Kaneda et al.
11. This work was extendedto multiple scattering by Nishita et al. 4. These methods require a lengthy simulationfor a given sky condition, but can work with arbitrarily complex atmosphericconditions. 4For simpler sky conditions, various researchers use parametric models. Pokrowskiproposed a formula for sky luminance based on theory and sky measurements.
Thiswas then improved by Kittler and adopted as a standard by the CIE 9. Furthermore,Perez et al. developed a five-parameter model describing the sky luminancedistribution 12. Although the Perez model is similar to the CIE model, it isslightly more accurate if the parameters are chosen well 4. Furthermore, an analyticalmodel that simulates actual observational data of daytime sky appearance ispresented by Preetham et al.
9.Less research has been conducted on the night sky and is limited mainly to aspectsof the night examined in isolation. For instance, Oberschelp and Hornug provideddiagrammatic visualizations of eclipses and planetary conjunction events 13. Theirfocus was not realistic rendering but the illustration of these events.
However, asystem to render the night sky was presented by Wann Jensen et al. and involves ageneral simulation of the night sky. 14Day and night sky simulations share several similarities, such as scattering of lightfrom the Sun and Moon, but some features that are only visible in the dimmer nightsky, for example stars, are ignored in daylight simulations. 14 Furthermore, theMoon is the main source of illumination in the night sky and due to its complexappearance, raises issues that the Sun does not. 14 Ideally, day and night skysimulation should not be viewed as two separate models, but rather a general modelthat combines both aspects.
Haber et al. bridge the gap between daytime and nightsky rendering 9. The presented simulation is based on the theory of light scatteringby small particles and models all effects that influence sky appearance when the Sunis both above and below the visible horizon 9.Two different approaches have been pursued for day-time sky rendering: modelsthat describe the sky appearance in terms of non-physical parameters and physicsbasedsystems that solve the 3D rendering equation and are based on actual atmosphericconditions 4. As previously mentioned, the aim of this project is to yieldphysically-accurate light probes, and this critically depends on the physically basedmodels used 8.
Hence a realistic model, that combines both day and night skies,and accurately simulates light interaction and effects that influence sky’s appearanceis essential for accurate light probes.