A Robust RGB-D SLAM with Loosely-Coupled Semi-Direct Method

In last few years, there has been growing interest in RGB-D Simultaneous localization and mapping (RGB-D SLAM) due to its ability of 3D dense reconstruction. Current main feature-based RGB-D SLAM methods are prone to fail in low-texture scenes due to insufficient feature matches. Previous studies of improving the robustness of RGB-D SLAM have focused on adding the line and plane features. However, these methods affect the real-time performance of the system. This paper aims to propose a real-time RGB-D SLAM method that has a robust performance in low-texture scene. We combined the strength of both direct methods and feature-based methods, which is the robust performance of direct methods in low-texture scenes and global optimization ability of feature-based method. The experimental results demonstrated that our method performs better than currently main RGB-D SLAM methods in terms of robustness and computational efficiency. Our method provides a prospective solution for RGB-D SLAM in low-texture scenes.