Irregular, Unknown Light Sources in Dynamic Global Illumination

Sharat Chandran
Computer Science and Engg. Dept.
IIT, Powai
Mumbai, INDIA 400 076
sharat@cse.iitb.ac.in
Mayur P. Srivastava
Computer Science and Engg. Dept.
IT BHU
Varanasi, INDIA 221 005
mayur_prakash@rediffmail.com

Abstract

The goal in global illumination solutions for dynamic environments is to update a scene based on past scenes. Current state of the art solutions are either not applicable, or unduly complex, when there are large changes in the illumination of unbounded number of objects. Such changes may be caused by the appearance of unexpected (at modeling time), irregular light sources.
Our solutions complements existing solutions, and are for the restricted subset of dynamic environments when new light sources appear, but the scene geometry does not change. We design and implement a simple solution based on hierarchical radiosity, and contrast it with an alternate solution.


Proposed Methods

For simplicity, the methods assume that the source of light appear on one of the preexisting patches.
This enables simpler, interactive algorithms, and at the same time provides solutions for a reasonable subset of situations in dynamic environments in global illumination. Our solutions work in the context of the hierarchical radiosity (HR) algorithm. The patches corespond to roots of quadtrees.
The algorithm refines user patches into sub patches and sets up a linear number of links.

Adaptive Refine (AR)
In the general case, the light source will intersect with some of the preexisting nodes of the HR algorithm. We therefore need to further refine nodes that would not have been expanded in the original algorithm.

Fractional Emissivity (FE)
Although adaptive refine is fast, it can take a large amount of time if the initial scene is dense, and the input light source is highly irregular and not aligned with the patch boundaries.
An even faster algorithm is possible if one accepts a coarser approximation of the light source.
The idea here is to mark those portions of the HR nodes that get affected by the light source as in adaptive refine (AR).
The departure with AR arises when AR decided to further refine a node n that the HR algorithm does not refine.
At this point, in the third algorithm, fractional emissivity (FE) is assigned to n based on the percentage of overlap between the light source and n.
This overlap is computed using an algorithm such as Hodgman-Sutherland.

Images from the Paper

Test Scene One (Cornell Room in side view) Test Scene Two (Office Scene) Test Scene Three (Cornell Room in Pigeon's View)




(a) Original Cornell room.


(b) Rendering using AR.
Figure 1: The Cornell room, assumed to have a window. The figure on the left shows the standard room lit by an interior light source. At a different point of day, the room is lit by an unexpected light source with arbitrary shape through the window. We would like to bind a solution based on the first one, instead of starting a new solution based on a new patch.




(a) An office with two flat screen monitors, presumably with screen savers on.


(b) The appearance changes because mouse motion causes the windows to break out the screen saver mode. Note that the positions of these new "light sources" is not known apriori.

Figure 2: Two different states of an office. The solution on the right is incrementally computed based on the illumination in the first by Algorithm AR.




(a) Original Cornell room.


(b) Rendering using AR.

Figure 3: Light filters through the southern wall onto the blue wall. The resulting scene is rendered incrementally in the first method on the right.




(a) Original Cornell room.


(b) Rendering using AR.

Figure 4: A second light appears on the eastern wall.




(a) Original Cornell room.


(b) Rendering using AR.

Figure 5: The light on the blue wall disappears to reflect the passage of time.




(a) Original Cornell room.


(b) Rendering using AR.

Figure 6: A wireframe is shown to visualize how adaptive refine(appearing on the right) further refines Previous Approach.