 |
 |
Accurate Light Source Acquisition and Rendering MPI Informatik, Saarbrücken, Germany |
|
 |
 |
 |
 |
Overview
Abstract
Realistic image synthesis requires both complex and realistic models of real-world light sources and efficient rendering algorithms to deal with them. In this paper, we describe a processing pipeline for dealing with complex light sources from acquisition to global illumination rendering. We carefully design optical filters to guarantee high precision measurements of real-world light sources. We discuss two practically feasible setups that allow us to measure light sources with different characteristics. Finally, we introduce an efficient importance-driven photon emission algorithm for our representation that can be used, for example, in conjunction with Photon Maps.Data Acquisition
We developed a measurement system that significantly improves on previous light source measurement approaches by projecting the light field emitted by the light source into a finite basis before sampling. This is done using a simple optical system. The shape and support of the basis functions are specifically designed for a particular sampling scheme. Based on the resulting measurements we can exactly reconstruct the least-squares approximation of the true light field in our basis. Alternatively, we can reconstruct with a more efficient, shift-invariant filter, and obtain a close approximation to the least-squares solution that is suitable for hardware accelerated rendering.The basic idea of our approach is depicted in the figure below. Light rays are emitted from the light source and hit a filter in a plane S. This plane is opaque except for the finite support area of the filter which can be moved to a number of discrete positions on a regular grid. The filter is a semi-transparent film, similar to a slide, containing the 2D image of a basis function. The light falling through this filter hits a second plane M, on which we are able to measure the irradiance with a high spatial resolution.
 |
Measurement Setup
Based on this principle, we have experimented with two basic experimental setups depicted below. In the first one (left side), we use the CCD sensor of a digital camera as the measurement plane M and replace the optical system of the camera with our filter. The second setup, to the right, requires more space and is harder to calibrate, but allows for measurements with a wider field of view.
|
|
Photographs of the two setups are shown below:
|
|
Rendering
For rendering, we attach the light field to a virtual geometric object, representing an impostor for the area from which light is emitted. This can be, for example, the glass surface of a car head light, or simply a planar polygon representing one of the light field planes. This geometry helps in positioning the light source in the 3D scene and can also be used to aid the sampling in our rendering algorithm.We then use the local energy in the light field (i.e. the size of the coefficients for the individual basis functions) as an importance function for shooting photons from the light source. The importance table can be pre-inverted, so that the location and direction of a new photon can be determined in constant time through a table lookup.
In our particular implementation we use this importance-driven photon emission algorithm together with photon maps, but the method can really be used with any algorithm that shoots particles from the light source, including bi-directional path tracing, Metropolis light transport, irradiance gradients etc.
Results
Here we show some results from a bike headlight. Results from a flashlight are shown at the top of this page. |
| Photo of a bike headlight |
 |
 |
Movies
 |
 |
 |
| simulated slide projector (inverted SIGGRAPH 2003 logo, 21x21 images) reconstruction of projector moving away from a wall (DivX encoded) |
acquired bike light (9x7 images) reconstruction of light moving away from a wall (DivX encoded) |
simulated slide projector (inverted SIGGRAPH 2003 logo, 21x21 images) reconstruction of projector moving away from a wall (MPEG encoded) |
Literature
- M. Goesele, X. Granier, W. Heidrich, H.-P.
Seidel: Accurate Light Source Acquisition and Rendering. Proc. of
SIGGRAPH '03 (Special issue of ACM Transactions on Graphics), 2003, pp 621-630.
(Slides from the talk.) - X. Granier, M. Goesele, W. Heidrich, H.-P.
Seidel: Interactive Visualization of Complex Real-World Light Sources. Proc. of
Pacific Graphics 2003, 2003. (accepted)
- W. Heidrich, M. Goesele: Image-Based
Measurement of Light Sources With Correct Filtering. UBC Technical
Report TR-2001-08.
- W. Heidrich, J. Kautz, Ph. Slusallek, H.-P. Seidel: Canned Lightsources. EG Rendering Workshop '98
