Dealing with degeneracy in essential matrix estimation

Estimation of 3-D egomotion from video input, also known as visual odometry, is an important issue for many applications today. Augmented reality (AR) and robotic systems for example rely heavily on correct pose and motion estimation. In this paper we discuss egomotion estimation from a single camera. We focus on the estimation of the essential matrix and problems which arise from degenerate configurations when using the well known normalized 8-point algorithm. Lately, the BEEM algorithm has been published, which is a combined approach of several RANSAC methods. It tries to guide essential matrix generation away from degenerate configurations. We argue, that there are still cases which are not covered by the BEEM approach and encourage the combination with an improved method for detecting degenerate configurations (IDD).