Estimation of Motion Parameters Using 2-D Lines without Correspondences Based on Virtual Electric Potential Model

Abstract

This paper addresses the estimation problem of the motion parameters using 2D lines without correspondences. The method is based on two main ideas. The first one consists to model the image as grid of the virtual electric potential and to exploit the corners detected by an improved version of Harris and Stephens detector. Characteristic of gradient vectors at the corners detected is used in order to draw the strait lines. The second one uses the invariance property of the correlation matrix eigenstructure decomposition. The correlation matrix is formulated from the directing vectors of the straight lines. To determine the correspondence between the lines and to remove the outliers we use an affinity function based on a heuristic criterion. The performances of the method are degraded considerably in presence of noise. For this reason, a preprocessing stage is suitable. It consists to estimate the noise correlation matrix by evaluating iteratively the noise subspace in order to improve the signal noise ratio (SNR). The robustness of the method to the noise and the outliers is remarkable in synthetic or real images.