Comparative study of SLAM systems in various environments using different sensing technologies
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3147965Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
- Master's theses (RealTek) [1826]
Sammendrag
This thesis explores a comparative study on camera based and LIDAR based Simultaneous Localizationand Mapping (SLAM) systems across diverse environments, specifically agricultural orchards and urbanparking lots. For camera based systems, it also includes indoor environments.
Traditional evaluations of SLAM technologies often rely on well-known datasets such as KITTI and TUM,which may not entirely reflect the operational challenges encountered in different real-world scenarios.Our study extends these evaluations to include real-world datasets collected from specific environmentsusing both a RGB-D camera and a LIDAR. This research utilizes a range of SLAM systems, includingORB-SLAM2, ManhattanSLAM, RESLAM, RGBDSLAMv2 and RTAB-Map for camera-based methods, as well as A-LOAM, F-LOAM, KISS-ICP, HDL Graph SLAM, SC-A-LOAM and LeGO-LOAM forLIDAR-based methods. We measured the accuracy of the pose estimation and the density of the resultingmaps of these systems, aiming to identify which technologies perform best under different environmentalconditions without adjusting the system parameters.
Initial findings indicate that environmental complexity significantly affects SLAM performance, withLIDAR-based systems showing greater robustness in less structured environments. This comprehensivecomparison not only aids in understanding the relative strengths and weaknesses of SLAM systems indifferent contexts, but also guides future enhancements in SLAM technology tailored to specific environmental challenges.