Motion and structure from the linear and non-linear algorithms

Abstract

This study presents the estimations of the 3D motion of a moving object in an image sequence taken from a monocular camera through linear and non-linear equations and determines the differences between linear and non[1]linear algorithms in terms of theoretical level and estimation accuracy with noisy point correspondences. Firstly, we investigated linear and non-linear algorithms for determining 3D motion at the theoretical level. Second, we estimated the 3D motion of the moving object in an image frame at two different instants of time with feature point correspondences in real-time. Finally, we implemented an accuracy analysis of the results from the linear and non-linear estimations. We showed that the non-linear approach produced more accurate results than the linear approach from noisy point correspondences.