180 likes | 294 Vues
This report explores a novel approach to improve monocular SLAM (Simultaneous Localization and Mapping) by integrating 3D city models, addressing the prevalent drift errors that hinder large-scale applications. The approach involves coarse SLAM reconstruction and bundle adjustment techniques, demonstrating its effectiveness through experimental results from synthetic and real sequences. The proposed method enhances reconstruction precision, paving the way for augmented reality applications where accurate localization is critical. The findings suggest that the integration of 3D models substantially improves SLAM performance.
E N D
Towards Geographical Referencing of Monocular SLAM Reconstruction Using 3D City Models: Application to Real-Time Accurate Vision-Based Localization Reporter : 鄒嘉恆 Date : 2010/05/04 CVPR ‘09
Outline • Motivation • Coarse SLAM Reconstruction • Bundle adjustment • Experimental results • Application • Conclusion
Motivation • Current SLAM methods are still prone to drift errors, which prevent their use in large-scale applications.
Coarse SLAM Reconstruction • SLAM(Simultaneous localization and mapping)
Coarse SLAM Reconstruction-SLAM reconstruction fragmentation • Use the idea suggested by Lowe in [11] to segment the trajectory. 3D points [11]D. G. Lowe. Three-dimensional object recognition from single two-dimensional images. Artificial Intelligence, 31(3):355–395, 1987.
Coarse SLAM Reconstruction-Non-rigid ICP • Goal: • Point-plane association : • Robust estimation : 3D Model
Coarse SLAM Reconstruction Result
Bundle adjustment • Advantages to use the barycentre of backprojections: • The movement of Qi’ is then directly linked to the displacement of the cameras. • Qi position is not used in the cost function.
Experimental results • Synthetic sequence
Experimental results • Real sequence • 640x480, 1500 meters
Conclusion • Proposed a new approach to correct large-scale SLAM reconstructions. • Proposed AR application shows that the obtained reconstruction precision is sufficient.