摘要
Simultaneous Localization and Mapping(SLAM) has broad applications such as driverless cars and indoor service robots. The SLAM techniques usually assume that environments are static, and it is difficult to obtain good accuracy in highly dynamic scenes. In this paper, we adopt the semantic segmentation method to filter out dynamic points according to prior knowledge. To avoid tracking failure due to insufficient feature points, epipolar geometry is used to retain potential dynamic points as much as possible. The proposed method is implemented based on the classical SLAM framework ORBSLAM2 and carried out experiments on the TUM RGBD datasets. Compared with ORBSLAM2, the localization accuracy of this algorithm
学科与关键词
关键词
SLAMSemantic SegmentationEpipolar Constraint
- Robust RGB: D-SLAM in highly dynamic environments based on probability observations and clustering optimizationLiu, Hailin; Tian, Liangfang*; Du, Qiliang; Xu, WenjieMeasurement Science and Technology, 2024, 35(3): 035405.
- RLD-SLAM: A Robust Lightweight VI-SLAM for Dynamic Environments Leveraging Semantics and Motion InformationZheng, Zengrui; Lin, Shifeng; Yang, Chenguang*IEEE Transactions on Industrial Electronics, 2024.
- Dynamic visual simultaneous localization and mapping based on semantic segmentation moduleJin, Jing*; Jiang, Xufeng; Yu, Chenhui; Zhao, Lingna; Tang, ZhenApplied Intelligence, 2023, 53(16): 19418-19432.被引 1
