Stability assessment and performance comparison of VSLAM frameworks on open indoor datasets

Abstract. This article examines the main directions of development of modern visual navigation technologies and highlights achievements and problems in this area. The article also considers popular VSLAM (Visual Simultaneous Localization and Mapping) frameworks and their main characteristics. Additionally, several indoor datasets are reviewed to highlight the importance of different testing environments when evaluating VSLAM frameworks. In the experimental part of the work, we compared three prominent VSLAM frameworks: ORB-SLAM 3, Basalt, and OpenVSLAM. For experimental study, 20 image sequences from the EuRoC and TUM-VI dataset were used. The study pays special attention to complex cases in indoor navigation, particularly those involving insufficient scene illumination, which poses significant challenges to the accuracy of VSLAM frameworks. The last part of this work provides a comparison of error estimates, including the absolute trajectory error and its variation throughout the entire estimated trajectory.