All Points Considered: A Maximum Likelihood Method for Motion Recovery
This paper addresses the problem of motion parameter recovery. A novel paradigm is offered to this problem, which computes a maximum likelihood (ML) estimate. The main novelty is that all domain-range point combinations are considered, as opposed to a single “optimal” combination. While this involves the optimization of nontrivial cost functions, the results are superior to those of the so-called algebraic and geometric methods, especially under the presence of strong noise, or when the measurement points approach a degenerate configuration.
KeywordsNoise Variance Scatter Diagram Geometric Method Algebraic Method Geometric Estimate
Unable to display preview. Download preview PDF.
- D. A. Forsyth, S. Ioffe, and J. Haddon. Bayesian structure from motion. In ICCV99, pages 660–665, 1999.Google Scholar
- R. Hartley and A. Zisserman. Multiple Views Geoemtry in Computer Vision. Cambridge University Press, 2000.Google Scholar
- K. Kanatani. Geometric computation for machine vision. In Oxford University Press, 1993.Google Scholar
- W. Press, B. Flannery, S. Teukolsky, and W. Vetterling. Numerical Recipes in C. Cambridge University Press, 1988.Google Scholar
- G. Speyer and M. Werman. Parameter estimates for a pencil of lines: Bounds and estimators. In ECCV, 2002.Google Scholar
- P. H. S. Torr and A. Zisserman. Concerning Bayesian motion segmentation, model averaging, matching and the trifocal tensor. In H. Burkharddt and B. Neumann, editors, ECCV98 Vol 1, pages 511–528. Springer, 1998.Google Scholar
- Z. Zhang, R. Deriche, O. Faugeras, and Q. Luong. A robust technique for matching two uncalibrated images through the recovery of unknown epipolar geometry. Artificial Inteligence, 78(1–2):88–119, 1995.Google Scholar