230 likes | 621 Views
Interactive 3D Modeling Using Only One Image Sujin Liu and Zhiyong Huang School of Computing 1. Introduction Problem: for VR systems, to create the models, usually with irregular shapes the current CAD modeling software addressed different problems
E N D
Interactive 3D Modeling Using Only One Image Sujin Liu and Zhiyong Huang School of Computing
1. Introduction • Problem: for VR systems, to create the models, usually with irregular shapes • the current CAD modeling software addressed different problems • Ideas: study the use of human interaction and one image • balance the interaction and automation VRST 2000
2. Related Work • Forward methods • CSG/B-Rep • Zeleznik et al.. SKETCH. SIGGRAPH 96 • Igarashi et al. Teddy. SIGGRAPH 99 • Implicit Surfaces • Shen and Thalmann. Metaball Human, Implicit Surfaces 95 • Production systems • Sakaguchi and Ohya. Botanical Tree. VRST 99 VRST 2000
Reverse methods • Computational Geometry based • Edelsbrunner and Mucke, 3D Alpha Shapes. TOG 94 • Hoppe et al. Curless and Levoy. Zero-set. SIGGRAPH 92, SIGGRAPH 96 • Amenta et al. Crust.SIGGRAPH 99 • Turner et al., Line Drawing Interpretation, VRST 99 • Model based • Thalmann and Thalmann. Human.IEEE CG&A 87 • Pighin et al. Face.SIGGRAPH 98 • Lee et al. Face.Eurographics 00 VRST 2000
Hybrid method • Debevec et al. Façade.SIGGRAPH 96 • Other one image based method • Beymer and Poggio. Face Recognition. ICCV 95 • Horry et al. TIP (Tour Into Picture).SIGGRAPH 97 VRST 2000
3. Our Work One 2D image Photogrammetric Modeling 3D model Human Interaction Texture Mapping VRST 2000
Photogrammetric Modeling • Purpose: to achieve automation by exploring the use of one image • Major steps: • Contour extraction • 2D skeleton computation • 2D meshing • 3D meshing • Texture mapping VRST 2000
Contour Extraction • Using the color clustering • the algorithm classifies the pixels into different clusters by comparing result of the color threshold of each cluster • Two clusters: foreground and background • the foreground is distinguishable from the background by colors • not necessary a pure color background VRST 2000
Example VRST 2000
2D Skeleton Computation • Purpose: to derive the skeleton of the 2D shape • The algorithm is based on the feature tracking and minimal spanning tree • KLT feature tracking: derive the feature points of the image • Minimal spanning tree: derive the skeleton of the 2D shape VRST 2000
Example VRST 2000
2D Meshing • Purpose: to derive a 2D mesh of the shape using the skeleton and contour • A variation of the constrained Delaunay triangulation, Qhull • http://www.geom.umn.edu/software/qhull/ VRST 2000
Example VRST 2000
3D Meshing • Purpose: to derive the 3D mesh as an initial shape of the model • Intuition: lift the 2D mesh with different heights for every vertices • Height is estimated by the color intensity • Requires human interaction most VRST 2000
Example VRST 2000
Two problems • Resolution decreases • similar to Teddy • Back meshing • human interaction VRST 2000
Texture Mapping • Straight forward for the front mesh • each vertex of the 3D mesh has its texture coordinate in photogrammetrc modeling • Problem for the back mesh • human interaction VRST 2000
Example VRST 2000
Other Human Interaction • Common to any modeling systems • picking, grouping, adding, deleting, displacing, etc. VRST 2000
Summary • A hybrid modeling framework • requires human interaction • has automations from the use of one image • not a stereo vision • not a model-based VRST 2000
4. More Experimental Results • Video VRST 2000
5. Conclusion and Future Work • We have proposed and implemented a hybrid modeling frame work using only one image • Future work: to address more general shapes VRST 2000
6. Acknowledgement • Dr. Leow Wee Kheng, Zhang Yong • NUS Academic Research Grant RP3982704 VRST 2000