TSG-SLAM: SLAM Employing Tight Coupling of Instance Segmentation and Geometric Constraints in Complex Dynamic...

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) systems face challenges in complex dynamic environments.
• Managing moving objects and environmental comprehension are key difficulties for current SLAM technologies.

Purpose of the Study:

• To develop a visual SLAM method tailored for complex dynamic environments.
• To enhance robot perception and understanding in dynamic scenarios.

Main Methods:

• A dynamic feature removal module is proposed, tightly coupling instance segmentation and multi-view geometric constraints (TSG).
• Instance segmentation identifies and removes dynamic features, preserving static features for fundamental matrix acquisition.
• Multi-view geometry constraints are used to identify and remove true dynamic features, maintaining static features for tracking and mapping.
• An instance-level semantic map is constructed for enhanced environmental understanding.

Main Results:

• The TSG-SLAM method demonstrates superior performance in detecting and eliminating dynamic feature points.
• The system achieves good localization accuracy in dynamic environments.
• Evaluated on TUM datasets and real-world scenarios, the proposed method shows effectiveness.

Conclusions:

• The TSG-SLAM approach offers a robust solution for visual SLAM in complex dynamic environments.
• Integrating semantic information with geometric constraints significantly improves SLAM performance.
• The method enhances robot capabilities in perceiving and navigating dynamic settings.