The EKF-based visual SLAM system with relative map orientation measurements

Abstract

This paper presents the extension of the feature-based, visual SLAM with additional measurements of the relative orientation between the current and past poses of the camera. The well known inverse depth representation of the point features was replaced with the combination of local maps and simplified features to allow orientation measurements via the estimation of the essential matrix. The proposed modification was evaluated using the openly available PUT RGB-D database. The incorporation of additional measurements resulted in reduction of the RMS of the trajectory reconstruction error by 17%.