BEBLID-SLAM: An Efficient Feature-Based Monocular SLAM System

The feature matching quality plays an important role in the robustness and location accuracy of feature-based Simultaneous Localization and Mapping (SLAM) system, in which descriptor is significant of tracking and re-localization. In this paper, we propose an efficient feature-based Monocular SLAM system, BEBLID-SLAM, which use BEBLID descriptor for feature matching in ORB-SLAM pipeline. In the proposed system, adopting histogram equalization to preprocess input images and offline training Bag-of-Words for BEBLID is respectively adopted to preprocess input images and realize re-localization and loop closure. Moreover, we valided that our algorithm has outstanding performance in robustness and accuracy than the popular algorithms in the public dataset EuRoC