Fig. 1. PC cluster for real-time active 3D object shape reconstruction.

Fig. 2. 3D video generation process.

Fig. 3. Silhouette volume intersection.

Fig. 4. 3D shape representations: (A) 3D voxel space, (B) parallel plane space.

Fig. 5. Plane-based volume intersection method.

Fig. 6. Linearized PPPP algorithm.

Fig. 7. Processing flow of the parallel pipelined 3D shape reconstruction.

Fig. 8. Average computation time for each pipeline stage.

Fig. 9. Computation time for reconstructing one 3D shape.

Fig. 10. Projection from an input image screen to the base plane.

Fig. 11. Size ratio between the projected silhouette and the observed silhouette.

Fig. 12. Camera Layout.

Fig. 13. Five different poses.

Fig. 14. Computation times of the base-plane projection by the original plane-based and 3-base-plane volume intersection methods.

Fig. 15. Frame and skin model.

Fig. 16. Photometric consistency and visibility.

Fig. 17. Silhouette preserving force.

Fig. 18. Camera arrangement.

Fig. 19. Initial shape. (A) is viewed from CAM₅ in Fig. 18, (B) from CAM₁.

Fig. 20. Experimental results. Top: (A) F_i(v) alone (α = 1.0, β = 0.0, γ = 0.0), Middle: (B) F_i(v) + F_s(v) (α = 0.5, β = 0.0, γ = 0.5) Bottom: (C) F_i(v) + F_e(v) + F_s(v) (α = 0.3, β = 0.4, γ = 0.3).

Fig. 21. Camera arrangement and synthesized object.

Fig. 22. Evaluating the effectiveness of the deformable mesh model. upper: super quadrics (n = 3.0, e = 1.0) with different degrees of surface textures: k% means k% of the surface area is covered with texture. lower: reconstructed object shapes.

Fig. 23. Reconstruction results for various n and e (with 100% textured surfaces).

Fig. 24. Roughly estimated 3D motion flow lines from t (in blue) to t + 1 (in red). The blue and red region denote the visual hull at t and t + 1, respectively (same as the center column of Fig. 25). (For interpretation of the references to colours in this figure legend, the reader is referred to the web version of this paper.)

Fig. 25. Successive deformation results (overview).

Fig. 26. Successive deformation results (detailed).

Fig. 27. Viewpoint independent patch-based method.

Fig. 28. Viewpoint and camera position.

Fig. 29. Depth buffer.

Fig. 30. Relations between patches.

Fig. 32. Images generated by the Viewpoint Independent Patch-Based Method.

Fig. 33. Images generated by the Viewpoint Dependent Vertex-Based Method.

Fig. 34. Sample images of generated 3D video with D-Mesh.

Fig. 31. Camera setting.

Fig. 35. Root-mean-square errors of RGB values (1).

Fig. 36. Subdivision of a surface patch.

Fig. 37. Mean side length in pixel on image planes of cameras.

Fig. 38. Root-mean-square errors of RGB values (2).

Fig. 39. Difference image between a real image and a generated image (frame #106).

Fig. 40. Example images visualized with original and subdivided patches (frame #103).

Fig. 41. Visualized 3D video with subdivided patches (frame#103).

View directions of the cameras: angles with YZ, XZ, and XY planes [DEG]