COMPUTER VISION AND VIRTUAL REALITY
ERCIM News No.31 - October 1997

BRIGHT ­ An Efficient Lighting Simulation System for Complex Scenes

by George Drettakis and François Sillion


Modern computer-based lighting systems are capable of providing realistic shadows and secondary illumination (multiple light bounces). Computer graphics images are thus less 'artificial looking' and much more appropriate to a vast number of applications such as architectural and design simulation, digital mock-ups, augmented reality, etc. At iMAGIS, a joint project between INRIA, the Centre National de la Recherches Scientifique (CNRS), Université Joseph Fourier and the Institut National Polytechnique de Grenoble, we have developed a prototype lighting simulation system called BRIGHT which is capable of providing lighting solutions for large input scenes (tens of thousands of polygons).

In the mid-eighties a new approach was introduced to simulate illumination for computer graphics using heat-transfer technology. This method is called the radiosity method, after the quantity describing light power per unit area leaving a surface. The images generated by radiosity methods have a much higher level of realism compared to traditional scan-conversion or ray-tracing approaches used previously in computer graphics. The input surfaces which typically come from a modeller, are subdivided into patches. This subdivision permits the visualisation of lighting variations and shadows (see Fig. 1a).


Figures 1a-d: Clustering in BRIGHT allows lighting of complex scenes and volumetric data.

One of the main problems with the original radiosity algorithm is the quadratic computation time in the number of patches. Intuitively, this is because the light interaction of every patch with every other patch has potentially to be considered. The hierarchical radiosity method provided an improvement over this approach, by representing light interactions hierarchically. Light transfers are thus performed at an appropriate hierarchical level of the subdivided patches, given a desired image quality. The initial cost, quadratic in the number of input objects, still remains however.

At iMAGIS, we have developed a new approach based on grouping (or clustering) of the original input surfaces. We initially create a hierarchy of clusters, and perform light transfers between clusters or between a cluster and a surface, depending on the precision required (see Fig 1b). This allows us to rapidly treat large scenes such as the one shown in Fig 1c. Using this approach BRIGHT can also treat scenes with volumetric data such as fire and smoke (Fig 1d ). Note how global effects (light on the ceiling and in regions of shadow) are accounted for in this image.

An important issue is the control of shadow precision. Given the cluster hierarchy, we can calculate shadows approximately, depending on the feature quality desired. For example, in Fig. 3, the user has specified that progressively finer detail is desired. We have developed a multi-resolution visibility algorithm which adapts the approximation to these needs.

Other recent research work at iMAGIS has included the treatment of scenes where objects move within the BRIGHT system. Traditional radiosity methods require that a full lighting solution be recomputed if an object is moved in a scene, even though only a small region of space is typically affected. By associating a shaft structure to each 'link' or light interaction, and by building a hierarchy of these shafts, it is possible to quickly identify the parts of the scene which have changed. As a result updates of the illumination in scenes can be achieved for tens of thousands of input objects at several frames per second.


Figure 2a-c: Progressively more detailed shadows are computed.

iMAGIS collaborates with the CIRAD in Montpellier on botanical applications using BRIGHT. The BRIGHT project is also used in the context of international collaboration with the University of Girona, the Fraunhofer Institute in Darmstadt, the University of Erlangen and the lighting software company LightWork Design.

The publication references for the above research can be found at: http://www-imagis.imag.fr/Publications/rendu.gb.html

More information can also be found at http://www-imagis.imag.fr/~Francois.Sillion.

Please contact:
George Drettakis or François Sillion - IMAG - INRIA
Tel: +33 4 76 63 56 46
E-mail: George.Drettakis@imag.fr or Francois.Sillion@imag.fr


return to the contents page