High-Precision Visual Localization and Dense Mapping Based on Visual SLAM for Indoor Environment

Abstract

Based on the situation of existing methods of indoor visual localization can hardly meet the fast, robust, practical and high-precision localization requirements simultaneously, a high-precision visual localization and dense mapping solution proposed in this paper. The solution possesses the capacity to estimate and optimize the 6-DoF pose and motion trajectory of a camera by adopting three parallel threads: tracking, local mapping and loop closing. Moreover, the method can reconstruct 3D dense map in real time by stitching point cloud module with input RGB-D images. Experiments show the proposed solution achieves excellent performance in terms of pose accuracy (centimeter) and localization speed (30FPS), which satisfies the requirements of indoor robot visual localization and 3D dense mapping with rapidity, preciseness and robustness.