Related Experiment Video
Updated: Feb 26, 2026
05:12
Robotized Testing of Camera Positions to Determine Ideal Configuration for Stereo 3D Visualization of Open-Heart Surgery
Published on: August 12, 2021
2.5K
Scale aware dense dynamic SLAM for monocular, stereo and RGBD cameras
Nuo Cen1, Yi Xu1, Tuck-Whye Wong1,2
1School of Aeronautics and Astronautics, Zhejiang University, Hangzhou, 310027, China.
Scientific Reports
|February 24, 2026
Summary
SDMFusion offers robust, real-time dense mapping for robots, even in dynamic environments. This scale-aware framework enhances autonomous navigation by accurately mapping static areas and rejecting dynamic objects.
Area of Science:
- Robotics
- Computer Vision
- Simultaneous Localization and Mapping (SLAM)
Background:
- Accurate real-time dense mapping is essential for autonomous robot navigation.
- Existing methods struggle with dynamic environments and often rely on single sensor types, limiting robustness.
Purpose of the Study:
- To propose SDMFusion, a generalized framework for real-time, scale-aware dense mapping with dynamic robustness.
- To support monocular, stereo, and RGB-D cameras within a unified system.
Main Methods:
- Integrating a scale-depth optimization module for absolute scale recovery and depth refinement.
- Implementing a dynamic feature rejection module using geometric constraints and motion consistency.
- Utilizing a real-time anti-dynamic reconstruction module for high-quality static map generation.
Main Results:
- SDMFusion demonstrated superior accuracy and robustness compared to ORB-SLAM3 and other dynamic SLAM methods.
- The framework effectively eliminates dynamic objects from dense maps.
- Validation was performed on multiple benchmark datasets (KITTI, TUM RGB-D, BONN RGB-D) and real-world scenarios.
Conclusions:
- SDMFusion provides a robust and accurate solution for real-time dense mapping in dynamic environments.
- The proposed framework enhances autonomous navigation capabilities by generating reliable maps of static surroundings.
