Visual and LiDAR SLAM Performance in Campus Scenes

Abstract

Simultaneous Localization And Mapping (SLAM) is a critical technology for environment perception in the fields of mobile robotics and autonomous vehicles, providing environmental information for robot system path planning and control. Depending on the primary sensors used, SLAM systems can be categorized into visual SLAM and laser SLAM. This paper collects datasets of Chongqing University and conducts comparative experiments by reproducing several state-of-the-art visual and laser SLAM algorithms. Based on the experimental results, it can be concluded that the integration of IMU signals in FAST-LIO can enhance the accuracy of map reconstruction, but it tends to experience drift during large-scale mapping. LeGO-LOAM, equipped with a loop closure detection module, demonstrates superior performance in overall trajectory accuracy compared to other algorithms. VINS-FUSION, by fusing a binocular camera and IMU, improves the global consistency and local precision of the map.