BY-SLAM: Dynamic Visual SLAM System Based on BEBLID and Semantic Information Extraction
View abstract on PubMed
Summary
This summary is machine-generated.This study introduces BY-SLAM, a dynamic visual simultaneous localization and mapping (SLAM) system. BY-SLAM effectively filters dynamic objects, significantly improving autonomous vehicle localization accuracy and map quality.
Area Of Science
- Robotics and Computer Vision
- Autonomous Systems
- Simultaneous Localization and Mapping (SLAM)
Background
- Traditional visual SLAM systems assume static environments, failing to account for dynamic objects.
- Dynamic objects in real-world scenarios degrade localization accuracy and can cause tracking failures in SLAM systems.
- Accurate localization and mapping are crucial for unmanned vehicle navigation.
Purpose Of The Study
- To develop a dynamic visual SLAM system, BY-SLAM, capable of handling dynamic targets.
- To enhance feature matching and semantic information extraction for robust SLAM.
- To improve localization accuracy and map quality in the presence of dynamic objects.
Main Methods
- Utilized BEBLID descriptor for Oriented FAST features to improve matching accuracy and speed.
- Employed FasterNet as the backbone for YOLOv8s to accelerate semantic extraction.
- Generated refined semantic masks using DBSCAN clustering for object detection.
- Filtered dynamic feature points using semantic masks and epipolar constraints.
- Constructed dense 3D maps excluding dynamic targets.
Main Results
- BY-SLAM effectively filters dynamic targets in both benchmark datasets and real-world scenarios.
- Achieved an average localization accuracy improvement of 95.53% on the TUM RGB-D dataset compared to ORB-SLAM3.
- Demonstrated superior localization accuracy, map readability, and robustness against classical dynamic SLAM systems.
Conclusions
- BY-SLAM offers a robust solution for dynamic visual SLAM by effectively handling environmental changes.
- The proposed system significantly enhances the performance of autonomous navigation in complex, dynamic environments.
- BY-SLAM provides a reliable foundation for precise positioning and mapping in unmanned vehicles.

