9.9 Three-Dimensional Reconstruction from Different Views

Chapter Contents (Back)
Includes both monocular cues and accumulation from stereo views, but it is feature based accumulation of the description. Fusion, Multiple Views. Shape from Monocular Cues. Multiple Views. Incremental Reconstruction.

9.9.1 MOSAIC System -- Herman

Chapter Contents (Back)
MOSAIC System. Multiple Views. Fusion, Multiple Views.

Herman, M.[Martin], Kanade, T.[Takeo],
Incremental Reconstruction of 3D Scenes from Multiple, Complex Images,
AI(30), No. 3, December 1986, pp. 289-341.
Elsevier DOI BibRef 8612
The 3D MOSAIC Scene Understanding System: Incremental Reconstruction of 3D Scenes from Complex Images,
DARPA84(137-148) BibRef
And: RCV87(471-482). BibRef CMU-CS-TR-84-102, CMU CS Dept., Feb, 1984. More on the implementation and the use of line junctions and the like. See the other papers. BibRef

Herman, M.[Martin], Kanade, T.[Takeo], Kuroe, S.[Shigeru],
Incremental Acquisition of a Three-Dimensional Scene Model from Images,
PAMI(6), No. 3, May, 1984, pp. 331-340. BibRef 8405 CMU-CS-TR-82-139, CMU CS Dept., October 1982. BibRef
And: DARPA82(179-192), Other version: BibRef
The 3D MOSAIC Scene Understanding System,
IJCAI83(1108-1112). Find the edges using a Sobel edge operator, link the edges together, generate straight line approximations, find junctions (T, Y, fork), match the junctions (and thus the lines), combine edges, generate the faces, get complete faces, find holes, etc. The claim is that this is one step beyond a stereo depth map (the scene specific knowledge of urban, buildings is used) it is a true 3-D compilation. Multiple views are used to fill in the gaps of earlier processing. One important point (not in the paper?) is to extract what you can at each image, do not try to force a decision, but wait until you can decide. (When in doubt, wait.) BibRef

Herman, M.,
Representation and Incremental Construction of a Three-Dimensional Scene Model,
T3DMP86(149-183). BibRef 8600

Walker, E.L., Herman, M., and Kanade, T.,
A Framework for Representing and Reasoning about Three-Dimensional Objects for Vision,
AIMag(9), No. 2, Summer 1988, pp. 47-58. BibRef 8800
And: ASR-I90(Chapter 6). BibRef
Earlier: SRMSF87(21-33). Reasoning on top of 3-D descriptions such as the earlier work. BibRef

Walker, E.L.[Ellen Lowenfeld], Herman, M.[Martin],
Geometric Reasoning for Constructing 3D Scene Descriptions from Images,
AI(37), No. 1-3, December, 1988, pp. 275-290.
Elsevier DOI four levels of representation: images, 2D features, 3D structures, and 3D geometric models. BibRef 8812

Walker, E.L.[Ellen Lowenfeld],
Knowledge-Based Image Understanding Using Incomplete and Generic Models,
IEEE DOI BibRef 9300

Herman, M.[Martin],
Matching Three-Dimensional Symbolic Descriptions Obtained from Multiple Views of a Scene,
CVPR85(585-590). Using range data and polyhedral representations. The match is based on junction types. BibRef 8500

Herman, M.[Martin],
Monocular Reconstruction of a Complex Urban Scene in the 3D MOSAIC System,
DARPA83(318-326). Similar to the above report (of October 82). BibRef 8310

Chapter on 3-D Shape from X -- Shading, Textures, Lasers, Structured Light, Focus, Line Drawings continues in
General Reconstructions .

Last update:Jun 23, 2018 at 14:58:54