Procedural Character Generation: Implementing Reference Fitting and Principal Components Analysis
Student thesis: Master Thesis and HD Thesis
- Andreas Madsen
- Esben Bach
4. term, Computer Science, Master (Master Programme)
This report examines the area of procedural generated content, more specifically
procedural generation of humanoid characters. The work presented extends upon an
analysis of which methods are suitable for generating such models. The analysis
suggest that procedurally generating humanoids is best done using principal
components analysis. Only the generation of the surface mesh is considered in
the analysis i.e. no texture data is generated. Based on the analysis we present
a design for an application which can be used by graphics artist in association
with their modeling editor. The application utilizes principal components
analysis, reference fitting and genetic algorithms to provide an easy method for
generating models. Based on the application implemented we conclude that it is
possible to generate humanoid models using principal components analysis,
however there is room for improvement. The test carried out do little to show
the degree of variation which can be achieved in the generation
process. Likewise the method developed for making the model generation intuitive
is not successful, given the small amount of test data.
Language | English |
---|---|
Publication date | Jun 2007 |