Title Page
Contents
ABSTRACT 10
Ⅰ. INTRODUCTION 12
Ⅱ. RELATED WORK 16
Ⅲ. PROPOSED METHOD 18
3.1. Linear RGB-D SLAM (L-SLAM) 18
3.2. Orthogonal Wall Detection using Four-Point LiDAR 20
3.2.1. Converting Range Measurements to 3D Point Cloud 20
3.2.2. Manhattan World Mapping in Real-time 24
3.3. Linear Four-Point LiDAR SLAM (FL-SLAM) 27
3.3.1. State Vector Definition 27
3.3.2. Propagation Step 28
3.3.3. Correction Step with the Global MW Map 28
Ⅳ. EVALUATION 33
4.1. Square Corridor 36
4.2. Open Hallway 1 36
4.3. U-shaped Corridor 37
4.4. Ablation Study on the Use of Four-Point LiDAR 37
Ⅴ. Conclusion 44
Ⅵ. REFERENCES 45
ABSTRACT IN KOREAN 49
Table. 1. Evaluation Results Of FDE(Unit: meter) on Author Collected Closed-Loop Datasets 35
Table. 2. Evaluataion Results Of ATE RMSE(Unit: meter) on Author-Collected Open-Loop Datasets 35
Figure 1. The accumulated 3D point cloud built using the four-point LiDAR and the results of the proposed method 15
Figure 2. The pipeline of the proposed FL-SLAM algorithm 22
Figure 3. The custom-built rig with four-point LiDAR and iPhone 12 Pro Max 22
Figure 4. The accumulated 3D poing cloud obtained with the four-point LiDAR and ARKit 6-DoF pose and the global MW map 23
Figure 5. Algorithm 1 - MW mapping with four-point LiDAR 26
Figure 6. Illustration of the Kalman filter components for the proposed method 31
Figure 7. Algorithm 2 - Step update in linear Kalman filter 32
Figure 8. Example images of open spaces of author-collected dataset 39
Figure 9. Selected 3D global MW map built with four-point LiDAR 39
Figure 10. Qualitative comparisons of the Square Corridor sequence 40
Figure 11. Qualitative comparisons of the Open Hallway 1 sequence 41
Figure 12. Qualitative comparisons of the U-shaped Corridor sequence 42
Figure 13. Ablation study on the use of mapping and localization with four-point LiDAR 43