Point-based modeling, animation and rendering of dynamic objects


Bart Adams

Katholieke Universiteit Leuven



Contact: Bart Adams

Ph.D. Thesis, Katholieke Universiteit Leuven, 205 + xiii p





Abstract

Dynamic objects are becoming ubiquitous in computer graphics applications. Algorithms, often based on physical laws, are proposed to alter the shape and appearance of virtual 3D objects. The visual richness of the resulting computer generated images has to match with the increasing demand for more and more realism. In this dissertation we assess a recent trend to resort to point primitives for the representation of such dynamic 3D objects and propose techniques to optimally exploit the characteristics of these point primitives in a wide range of computer graphics applications. It is shown how the absence of explicit connectivity information between the point samples significantly facilitates the handling of dynamically changing objects. Based on customized data structures and resampling operators, efficient algorithms are proposed for point-based modeling, animation and rendering of dynamic objects.

In the first part of this thesis, geometric modeling algorithms are proposed to interactively alter the shape and appearance of point-sampled 3D surfaces. Building on efficient point data structures, we show how boolean operations such as the union and intersection of complex surfaces can be computed at interactive rates. A 3D painting application is proposed entirely centered around the use of point primitives. It allows painting arbitrary levels of detail on point-sampled surfaces using virtual 3D brushes. Local resampling operators are given to maintain an adequate sampling density during the whole modeling process.

The second part of this work uses point primitives for physics-based modeling of dynamic objects. Next to point sampling the object's surface, a point-sampled volume representation is used to efficiently solve the equations governing the dynamic behavior. By decoupling the volume and surface representation, a trade-off can be made between visual complexity and physical accuracy. We show how the use of point primitives leads to efficient algorithms for the simulation of elastic solids and viscous fluids and propose dynamic point sampling operators to handle brittle and ductile fracturing.

The last part of this dissertation discusses ray tracing algorithms for point set surfaces. Spatial and temporal coherence is exploited to efficiently visualize dynamically changing point surfaces. Sharp features, such as the ones obtained by boolean operations and fracture animation, are faithfully reproduced using the concept of surface clipping relations. The discussed rendering algorithm is illustrated on various examples generated with the point-based animation framework introduced earlier in this dissertation.

All presented techniques contribute to the increasing requirement to efficiently handle dynamically changing objects in computer graphics applications.




Downloads


Links