IMU aided RGB-D SLAM

In recent years, a low-cost range sensing technology (such as Kinect) has attracted many researchers in vision community. Despite being promising for indoor navigation, continuous operation in a challenging environments is still a fundamental problem. This paper presents a real-time approach of combining range-color (RGB-D) cues with inertial information in a loosely-coupled framework for Simultaneous Localization and Mapping (SLAM). The method handles the degenerate cases arises in the RGB-D approaches i.e limited range of Kinect (approx. 4m), texture-less/planar environmental ambiguities. Experimental results are provided for indoor lab environment and evaluated against a centimeter-level accurate ground truth data. We showed that the proposed approach significantly reduces error in RGB-D pose estimation by bounding the errors implicitly by the inertial gravity vectors and accelerometer data.