RTS-SLAM: A Trajectory Consistency-Driven Multi-Constraint Dynamic Feature-Rejection Method for Visual SLAM in...

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous navigation.
• Traditional SLAM methods struggle in dynamic environments due to reliance on environmental rigidity.
• Dynamic environments cause significant accuracy degradation in conventional SLAM systems.

Purpose of the Study:

• To develop RTS-SLAM, a real-time semantic visual SLAM system specifically for dynamic environments.
• To enhance localization accuracy and mapping efficiency in the presence of environmental changes.
• To overcome the limitations of existing SLAM approaches in non-static settings.

Main Methods:

• RTS-SLAM builds upon the ORB-SLAM2 framework, incorporating a multi-layer, constraint-driven dynamic feature-rejection strategy.
• Dynamic features are removed using semantic information and geometric constraints, followed by trajectory-consistency analysis.
• A dense mapping strategy with global sparsification and critical region refinement reduces memory usage while preserving essential geometries.

Main Results:

• RTS-SLAM demonstrated a reduction in average absolute trajectory error by over 95% compared to ORB-SLAM2 on dynamic datasets (TUM RGB-D and Bonn).
• The system maintains real-time performance, crucial for practical autonomous navigation.
• High localization accuracy was achieved even in challenging dynamic environments.

Conclusions:

• RTS-SLAM effectively addresses the challenges of SLAM in dynamic environments.
• The proposed feature rejection and mapping strategies significantly improve accuracy and efficiency.
• RTS-SLAM offers a robust solution for real-time autonomous navigation in complex, changing surroundings.