Conventional SLAM only requires the construction of sparse maps for localization, while in order to meet the safe driving needs of unmanned vehicles, they need to understand the edges of the road, i.e., with a semantic level of understanding. In addition, unmanned vehicles are more sensitive to lateral errors than longitudinal errors, which requires SLAM algorithms with higher accuracy for lateral errors. We investigate the ORB-SLAM3 algorithm by introducing satellite maps as a priori knowledge, using the corners in satellite maps to initialize the odometer, remove the accumulated errors, and correct the previous positions; using the results of particle filtering and lane line identification to further optimize the localization results of ORB-SLAM3 and to draw maps with lane semantic information. Our experiments show that our algorithm significantly reduces the cumulative error without loopback, improves the localization accuracy, and yields lane line maps with large engineering applications.
Visual localization and mapping based on road constraints
3rd International Conference on Internet of Things and Smart City (IoTSC 2023) ; 2023 ; Chongqing, China
Proc. SPIE ; 12708
2023-06-14
Conference paper
Electronic Resource
English
MONOCULAR VISUAL LOCALIZATION USING ROAD STRUCTURAL FEATURES
British Library Conference Proceedings | 2014
|A Visual-Inertial Localization Method Based on Motion Feature Constraints
Springer Verlag | 2025
|SIMULTANEOUS LOCALIZATION AND MAPPING USING ROAD SURFACE DATA
European Patent Office | 2022
|Synchronous localization and mapping using road surface data
European Patent Office | 2023
|