DualViewSemanticSLAM is a Visual Simultaneous Localization and Mapping (SLAM) system that integrates both bird's eye and front views for enhanced autonomous navigation and mapping capabilities.
- Initial Mono Performance: The system began with a strong performance using only the front mono view.
- Enhanced Point Collection: Modifications were made to collect more points from the bird's eye view by not applying certain masks.
- Motion Model Prior: Incorporation of a motion model prior to reduce the number of lost matches in experiments.
- Point Integration: Both front and bird's eye view points are utilized for a more robust mapping.
- Pose Graph Optimization: Implementation of pose graph optimization in local mapping and global bundle adjustment to maintain tracking stability.
- Re-initialization Strategies: Two re-initialization methods were introduced to recover from lost tracking, with the second method showing successful results.