Towards Resource-Aware Visual Inertial SLAM
Abstract
Simultaneous Localization and Mapping is a critical component in Augmented Reality applications with Smart eyewear devices. Despite advances in visual SLAM, recent research has prioritized precision over efficiency. Our work emphasizes the efficiency-accuracy trade-off in a particular SLAM system, namely ORB-SLAM3. By testing different sensor setups and configurations, we found the stereo-inertial setup to be the best choice for the trade-off. Furthermore, we propose a resource-aware modification to ORB-SLAM3, scaling Bundle Adjustment complexity based on computation time, enhancing real-time performance without a significant accuracy loss. And additionally, in some cases it resulted even in an increase of accuracy, opening new promising directions for Bundle Adjustment optimization.
Cite
Text
Affatato et al. "Towards Resource-Aware Visual Inertial SLAM." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91989-3_15Markdown
[Affatato et al. "Towards Resource-Aware Visual Inertial SLAM." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/affatato2024eccvw-resourceaware/) doi:10.1007/978-3-031-91989-3_15BibTeX
@inproceedings{affatato2024eccvw-resourceaware,
title = {{Towards Resource-Aware Visual Inertial SLAM}},
author = {Affatato, Giovanni and Paracchini, Marco and Palermo, Francesca and Trojaniello, Diana and Ongarello, Tommaso and Marcon, Marco and Tubaro, Stefano},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {228-245},
doi = {10.1007/978-3-031-91989-3_15},
url = {https://mlanthology.org/eccvw/2024/affatato2024eccvw-resourceaware/}
}