3-D Reconstruction of Three Views Based on Manifold Study - Advances in Image and Graphics Technologies - page 190

a b
Fig. 10. a. Result of dimensionality reduction using ISOMAP and b. subdivision of plane
Delaunay triangle nets
Fig. 11. 3-D TIN and reconstruction result with texture mapping
In conclusion, the 3-D reconstruction method proposed in this article, which based
on manifold study by unit of three views, achieves the following optimizations:
(1) With no computation of trifocal tensor, this methods obtains precise calculation of
the fundamental matrix relying on the optimized inliers set, getting rid of mismatches
by three views constraint, which could be detected just by epipolar geometry.
Meanwhile, more 3-D points are obtaining than two views. It overcomes the
insufficient that the number of 3-D points declines rapidly during adding the distance
between adjacent views to some extent. Consequently, this method expands the range
of 3-D reconstruction applications based on image sequences.
(2) To exclude the outliers in reconstructed 3-D points, the theory of minimum
spanning tree is applied. Then the accurate initial point cloud is obtained, which is the
basic of 3-D topological reconstruction.
(3) The initial 3-D points represent the obvious feature points of structure or texture in
actual scene, application of manifold study could make sure that those points must be
on the reconstructed scene. what's more, the correctness of topological relations is also
be guaranteed.
References
1. Armangue, X., Salvi, J.: Overall view regarding fundamental matrix estimation. Image and
Vision Computing 21(2), 200-205 (2003)
2. Zhong, X. H.: Research on methods for estimating the fundamental matrix. Jilin University
(2005) ( 基本矩阵计算方法的研究 )