Point-based rendering is an active research area, driven by the need to handle the vast quantities of data derived from object scanning.

Conventional polygon-based rendering in current graphics cards is inefficient since typical polygons may generate only a handful of pixels and the cost of setting up polygon shading parameters is wasted if no interior pixels are produced. Thus point-based rendering proceeds direct to pixels without an intermediate approximation phase.

Currently graphics cards are increasing in capacity at a rate greater than Moore's Law and given the increased complexity of typical models and the relatively static pixel count of output devices, the case for point-based rendering is obvious. In the case of geometric modelling, different geometric primitives are needed for different tasks and melding them together in a single system raises many issues. Geometric computations are notoriously fraught with difficulties, both numerical and geometric.

Robust algorithmic generation of point sets based on regular integer grids and algorithms for the combination of such point sets may be a better proposition than handling the original geometry by polygonal approximation using floating point arithmetic. For this the goal is to develop a new geometry which is mathematically sound, amenable to computation, and for which proofs of correctness of algorithms can be derived, in short a truly computational geometry.

Research Team

Prof. A.R. Forrest

Collaborators

Prof. Q.S. Peng of the State Key Laboratory of Computer Aided Design and Computer Graphics, Zhejing University, Hangzhou and Prof. A.E. Fabris of the University of Sao Paulo, Brazil as well as Dr. M.A. Sabin in Cambridge. Funding: Royal Society/National Science Foundation of China.