Semantic-Structure-Aware Multi-Level Information Fusion for Robust Global Orientation Optimization of Autonomous...

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous mobile robots in unknown environments.
• Conventional SLAM methods struggle with environmental interferences due to limited use of semantic information.
• Existing data-driven feature extraction lacks robustness against environmental uncertainties.

Purpose of the Study:

• To develop a robust global orientation estimation scheme for multi-camera SLAM systems.
• To enhance SLAM performance by integrating rich semantic information from the environment.
• To improve the accuracy and reliability of localization for mobile robots.

Main Methods:

• A visual semantic perception system using a synthesized surround view image from multi-eye cameras.
• Semantic segmentation to extract passable space and create a mask for feature extraction.
• Hybrid edge information extraction, utilizing reverse perspective projection distortion characteristics to eliminate edge distortions.
• Multi-level information fusion for robust global orientation estimation.

Main Results:

• Achieved at least a 10% improvement over existing edge segmentation methods.
• Demonstrated significantly reduced orientation estimation errors (1-2 degrees).
• Showcased substantially smaller trajectory drift values compared to other methods.
• Validated performance on an intelligent vehicle for automated valet parking in diverse scenes.

Conclusions:

• The proposed semantic-aware multi-level information fusion scheme significantly enhances SLAM robustness and localization accuracy.
• Integrating semantic information and refined edge extraction leads to superior orientation estimation.
• The method offers a practical solution for reliable autonomous navigation in complex indoor and outdoor environments.