About the Project
Lighting design is of high importance in pipelines for the creation of three-dimensional environments for cinematic computer graphics, video games and architectural design, including the energy-efficient design of buildings. Typically, this is an iterative manual process involving the placement and configuration of luminaries and the modification of geometry, in a trial-and-error basis. The complexity of the process has led to the research of automatic methods for determining the optimal configuration of luminaries (Inverse Lighting Problem - ILP), and geometry (Inverse Geometry Problem - IGP), given a set of specific illumination goals and constraints. The Opening Design constitutes one of the most important sub-problems of the IGP, aiming at the determination of openings on the geometry, such as windows or skylights, in order to optimize the contribution of the environment lighting to the illumination of the 3D environment.
Based on the open problems and restrictions of each application area, the common underlying theory, the shared demand for interactivity and the advances in GPU-based global-illumination techniques, GLIDE aimed at integrating the lighting design early on in the modelling process, so that interactive goal-driven lighting and geometry editing can happen seamlessly and simultaneously, for full-scale dynamic environments and with no pre-compotation steps, thus offering an entire new range of creative possibilities and tools to the artist, architect and designer. To this end, in the GLIDE project we attempted to address the two major sub-problems of inverse rendering, i.e. IGP and ILP, by discovering new, efficient and more accurate ways to model and solve them, while dispensing with any pre-computation and view-dependency constraints.
For the inverse lighting problem, we investigated a general and automatic approach, which enables the rapid positioning and emittance estimation of point light sources in arbitrary 3D environments and results in plausible and efficient lighting solutions, given the desired target illumination at arbitrary locations in the environment. Furthermore, additional user constraints can limit the placement of light sources only at specific distances from the geometry or at given partitions of the 3D model and even impose energy minimization or lighting uniformity, rendering the approach ideal for energy-efficient architectural design. To this end, we employed a special hierarchical light clustering that overcomes limitations of previous approaches in environments with high occlusion or structural complexity.
For the problem of opening design, a novel method was devised that determines the number, location and shape of the openings, given the designer's desired illumination at various locations in a 3D environment as well as other constraints such as atmospheric conditions and time intervals of interest, or which surfaces are candidates for bearing openings. The desired illumination can be provided as specific irradiance levels or uniform lighting within certain irradiance bounds. The method operates by virtually "dicing" the candidate surfaces for openings in a non-destructive manner and optimizing the contribution of each one of the discrete elements. The results are further consolidated to form architecturally plausible (e.g. rectangular) openings. Due to the generic and physically-correct underlying lighting evaluation processes, the natural constraints and the inclusion of realistic outdoor lighting conditions, our method is suitable for the design of energy-efficient buildings.
Finally, for the enabling technology behind the scenes, we investigated fast hardware-accelerated approaches for the physically-correct energy transport and lighting evaluation, in order to bring the heavy computations for the lighting optimization in both problems to practical execution times. To this end, we investigated both established parallel GPU-based techniques and novel rasterization-based ray tracing solutions that bring the full evaluation of lighting at thousands of sampling locations to mere milliseconds, for photorealistic results. In particular, since both the configuration of light sources and the opening design problem imply dynamic geometry and 3D environment setup, we focused our research on the creation of ray tracing techniques for fully dynamic scenes, where every aspect of the environment is constantly re-configured.
GLIDE has been co-financed by the European Union (European Social Fund - ESF) and Greek national funds through the Operational Program ”Education and Lifelong Learning” of the National Strategic Reference Framework (NSRF) - Research Funding Program: ARISTEIA II (grant no.3712). The project ran from February 2014 till October 2015 and resulted in 7 research publications and 1 poster presentation. Preliminary results of our rendering algorithms have been demonstrated at conferences and other public venues and the methodology of the project has been disseminated both at scientific communities and wider public. Demonstration software, videos and source code for various components are also publicly available through the project site.
Contact info: Georgios Papaioannou, gepap @ aueb gr