Research On Diversity Generating Method Of Virtual Crowd

Crowd simulation has been applied in a wide variety of areas, ranging frommovie animation, urban planning and training simulations to games. In crowdsimulation, if all of the virtual characters look the same, it will appear to be clonesand lack feeling of reality as in real world. The thesis studies the diversifiedappearance generating method of characters from aspect of mesh models, clothtextures, and animation postures. Character model difference is realized by usingHasler’s model, various posture is obtained by skeleton embedding technology, anddiversification of the appearance is achieved by texture template method.The main work of this thesis is listed as follow:(1) A diversified model generating method of characters is put forward basedon the available database of Hasler and automatic rigging technique ofBaran. Firstly, semantic information is used for morphing models andgenerating various instances by Hasler’s method. Secondly, the skeleton isrigged to the diversified character by Baran’s method and attached toskinning by heat equilibrium theory, and then animation is performed byimporting the motion data. The fine combination of two methodseffectively generates diversified crowd models and animation.(2) A method utilizing texture templates is proposed to achieve diversificationof character appearance. UV is unfolded for the mesh model withouttexture coordinates as texture templates, which are processed by imagesegmentation algorithm to obtain semantic information of clothing, andfilled with pure color or image tiles to produce various textures andachieve appearance diversity in crowd simulation.(3) Prototype system platform is designed to conduct experiments and analysis.Results show that the proposed method can generate plausible virtualcrowds with good diversity.