Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
527313 | Image and Vision Computing | 2010 | 9 Pages |
This paper shows that we can classify latent outliers efficiently through the process of minimizing the sum of infeasibilities (SOI). The SOI minimization has been developed in the area of convex optimization to find an initial solution, solve a feasibility problem, or check out some inconsistent constraints. It was also adopted recently as an approximation method to minimize a robust error function under the framework of the L∞L∞ norm minimization for geometric vision problems. In this paper, we show that the SOI minimization is practically effective in collecting outliers when it is applied to geometric vision problems. In particular, this method is useful in structure and motion reconstruction where methods such as RANSAC are not applicable. We demonstrate the effectiveness of the method through experiments with synthetic and real data sets.