Presented by Konstantinos Georgiadis

Transcription

1 Presented by Konstantinos Georgiadis

2 Abstract This method extends the Hierarchical Radiosity approach for environments whose geometry and surface attributes changes dynamically. Major contributions It maintains the accuracy of the mesh It handles changes in geometry

3 Previous Work Radiosity View independent, but still only for static environments Progressive Radiosity Hierarchical Radiosity No mechanism for mesh unrefinement because: It is impossible to maintain the mesh history No control over accuracy Handles refinement/unrefinement of the mesh but: The accuracy of the mesh is not maintained It doesn t handle changes in geometry

4 Dynamic Environments Surface Attribute Manipulation Changes in the light source or in the reflectance of a surface Geometry not effected. Solution Adjust the mesh to maintain accuracy Geometry Manipulation Changes in geometry affect the form factors Still, most links remain unchanged Solution A taxonomy of links reduces form factors recomputation

5 Taxonomy of Links Scene surfaces are divided into two groups For example, a chair.

6 Taxonomy of Links The links are classified into three categories Links can be: Updated Deleted Created Or remain unchanged

7 Object - Object In few occasions, rotation may cause self occlusion. The modification of object-object links is of moderate complexity. Remain unchanged Become occluded Delete link Become visible Create link

8 Object - Environment The modification of such links is of easy complexity, because the dynamic object is always known. Visibility unchanged Update link Become occluded Delete link Become visible Create link

9 Environment - Environment The modification of such links is of hard complexity. An exhaustive search must be performed to identify them. Visibility unchanged Become occluded Delete link Become visible Create link

10 Mesh Optimization The unrefinement of the mesh is called mesh folding and it introduces two new interactions : Ghost links Shadow links The mesh (un)refinement depends on:

11 Motivation Changes in the emissive power of the light sources affects the brightness of the scene. If brightness : Increase Mesh refinement Decrease Mesh unrefinement? Solution Use Ghost links The reposition of objects leaves under-refined and overly refined mesh elements due to occlusion. Under-refined areas Automatically meshed Overly refined areas How to remove them? Solution Use Shadow links

12 Ghost Link Creation Ghost links are created during the refinement process. Instead of discarding information from energy links, store it in ghost links. No further computational cost Storage cost moderate Information retained by a Ghost link Form factor Visibility information

13 Shadow Link Creation Shadow links are created during refinement process, when partial visibility causes sufficient decrease in energy to terminate subdivision. The number of shadow links is optimal Information retained by a Shadow link Receiver patch Source patch A set of occluding objects Form factor

14 Global Mesh Folding with Ghost Links Energy delta Difference between new/old attribute values. Computed for each color channel. Positive delta Patch refining (Energy increase) Negative delta Patch folding (Energy decrease) Conservative delta Ensure optimal mesh by invoking folding for any negative delta

15 Global Mesh Folding with Ghost Links Mesh folding Performed in two steps: Step 1: Delete links When energy is decreased, ghost links become energy links. Energy links between patch children get deleted. Step 2: Delete patches Patches with no children than do not contribute energy to the scene get deleted. Accomplished through use count.

16 Local Mesh Folding with Shadow Links Local mesh folding similar to global mesh folding; only shadow links are used instead of ghost links. Shadow clean-up Solution Recompute the form factors of the energy links from the light source to the receiver patches of the shadow links.

17 Results Ghost links Original Image Middle Image Final Image Shadow links Original Image Middle Image Final Image

18 Scene Partitioning The objects of a dynamic environment are organized by using: Hierarchical Bounding Volume (HBV) Object Clustering Furthermore, the objects can be manipulated by using: Motion Volume

19 Hierarchical Bounding Volume A hierarchical bounding volume (HBV) is a data structure that groups individual bounding volumes into larger and larger volumes. This method uses an automated HBV in order to : Remove a bounding volume subtree from hierarchy Add and remove clustered objects

20 Object Clustering Object clustering defines a class of movable objects and aids a ray-acceleration scheme which is important for visibility test. The sub-objects of all dynamic objects are clustered together into sub-hbv. When selected the clustered object is removed from scene When released it is transformed and re-inserted back into the scene Scene modeled as an n-tree HBV built from this tree

21 The Motion Volume The motion volume is a scene-partitioning bounding volume which encloses the original and the repositioned object. It consists of a collection of plane-sets that enclose the range of motion of the object

22 Creating the Volume The motion volume is created by using: the bounding box of the dynamic object the transformation matrix that takes the object to its new position The eight vertices of the original bounding box transform to eight new vertices. An enclosure around the sixteen points yields the motion volume.

23 Object Clipping Cull links that have not been affected by an object's motion using: The motion volume A modified 3D Cohen-Sutherland clipping algorithm Methodology: Check the outcodes of pairs of objects to decide if their links can be excluded from the search. Reject objects that lie on the same side of a motion volume plane.

24 Object Visibility The remaining pairs of clustered objects will either be: mutually visible (links may exist) completely occluded (no link exist) The outcode test is irrelevant to object-to-object visibility, thus a visibility test is required.

25 Surface Visibility Links between visible objects are limited to particular surfaces. Some surfaces can be rejected based on their orientation. Method Determine if two surfaces are visible Examine the surfaces of each object For every surface normal N, examine the projection onto N of the other object's maximum bounding vertex in the direction of N. If the projection lies in the positive half space of the surface then surfaces are visible and their links must be recomputed.

26 Updating the Links

27 Advantages - Drawbacks - New mesh folding methods - Novel scene-partitioning technique to speed the update of a system - The update time for most scenes is still too slow to be interactive. - The modification and creation of objectobject links are naively solved

28 Future Work Object clustering and lazy evaluation should be investigated for the modification of object-object links and the creation of new links. Incorporating new techniques for visibility preprocessing or meshing along lines of radiance discontinuities may speed up computations. Utrecht University - Advanced Graphics

29 Questions? Thank you for your attention