3D-Modeling by Ortho-Image Generation from Image Sequences

1. 3D-Modeling by Ortho-Image Generation from Image Sequences Thorsten Thorm¨ahlen Hans-Peter Seidel Max Planck Institute for Computer Science∗, Saarbr¨ucken, Germany ACM SIGGRAPH 2008

2. Introduction • Fast interactive generation of 3D models from image sequences • Generate high-quality 3D model of a static object from an image sequence • It canbe generated for translucent or specular surfaces

3. Introduction This paper

4. Outline • Introduction • Placement of the Bounding box • Ortho-image Generation • Ortho-images without geometry • Ortho-images with approximate geometry • Results • Conclusion

5. Placement of the Bounding box • 4 × 4 matrix H describes the anisotropic similarity transformation, which has 9 parameters, 3 each for position, rotation, and anisotropic scaling.

6. Ortho-images without geometry • If many camera views are available • Determine the colorintensity for that pixel directly if min{γCj} < τ=2◦ .

7. Ortho-images with approximate geometry • If manymin{γCj} > τ • Then the color intensity of the pixel is determined by projecting a ray from the camera center Cj to the point • But we need approximate geometry now!!

8. Ortho-images with approximate geometry • Graph-cut approach for multi-view stereo in [Vogiatzis et al. 2005] • For each of the voxels, a photo-consistency score is calculated • The graph-cut algorithm finds a cut through the edges of the graph

9. Result video display

10. Conclusion • Semi-automatic approach • High-quality and accurate 3D models from image sequences • Allow specular or transparent objects