Orientation estimation via low cost depth sensor ICP versus MEMS gyroscope sensor fusion

Abstract

Working implementations of 3D simultaneous localization and mapping (SLAM) using low cost depth sensors have exploded in popularity in recent years but remain limited in some important ways. In particular, they are not robust to rapid changes in orientation and can accumulate significant error with just a single gradual turn into a new scene. In this work we integrate and compare the Kinfu iterative closest point (ICP) based SLAM implementation from the Point Cloud Library with a hybrid optical-based inertial tracker (HObIT). In three separate experiments we find the HObIT to be far more accurate and robust to both slow and rapid changes in orientation. We therefore propose the integration of precision calibrated MEMS inertial sensors into existing low cost SLAM solutions for far more practical and robust solutions.