M. Trujillo and E. Izquierdo (UK)
Fundamental matrix, clustering, disparity, Least Median of Squares.
In this paper an approach to estimate the fundamental matrix is proposed. The goal is to overcome the high vulnerability of linear models against noise and bad estimates by exploiting the structure of the input data. Initially, a small number of low-dimensionality least square problems are solved using well-selected subsets from the input data. The selection process is based on the inherent 3D structure encapsulated in the disparity vectors. It is shown that the 3D structure embedded in the input data provides means to filter redundant information and to obtain better estimates with few input points. The results of these estimations are fed into a Least Median of Squares schema, which is applied to recover the final estimate of the Fundamental Matrix. Several experiments were conducted to assess the performance of the proposed technique.
Important Links:
Go Back
