Recovery of egomotion from optical flow with large motion based on subspace method

In this paper a novel algorithm to determine optical flow field with large motion is presented. Then, a subspace method for estimating 3D egomotion from optical flow computed is proposed. The algorithm of optical flow is robust and reliable to the large motion, which is very important to recover the 3D egomotion parameters. Based on the flow velocity computed from image sequence, the subspace method is carried out in three steps using three sets of equations derived from nonlinear equation of motion perspective. At first, the translational direction of the observer's motion is recovered by searching a candidate over a discrete space to minimize a residual function. Once the translation has been estimated, the rotation components of the observer's motion can been resolved from the second set of equations by using the least square optimization. At last, the relative depth map of the scene is recovered using the third set of equations. Promising quantitative results are reported from experiments with simulated data and synthetic image.