A lightweight approach to 6-DOF plane-based egomotion estimation using inverse depth

Abstract

Egomotion estimation is a fundamental building block for self-localisation and SLAM, especially when an accurate map is not available to the robot. This paper presents a real time plane-based approach to estimate the 6 degrees of freedom (DOF) pose of a moving Microsoft Kinect. An inverse depth formulation is used to robustly t multiple planes in the environment, which are then used to provide an egomotion estimate. Experiments presented demonstrate the algorithm’s accuracy, repeatability and ability to handle dierent motions and scene compositions. Despite using full resolution Kinect depth images, our approach runs at 10-13 frames per second.