Geometric constraints and semantic optimization SLAM algorithm for dynamic scenarios.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional visual SLAM (simultaneous localization and mapping) systems struggle in dynamic environments, leading to errors and redundancy.
• Existing methods often fail to accurately track features in scenes with moving objects.

Purpose of the Study:

• To develop a dynamic feature detection and filtering algorithm for robust visual SLAM.
• To enhance the adaptability and accuracy of SLAM systems in complex, dynamic environments.

Main Methods:

• A feature point selection and optimization strategy using quadtree nodes prioritizes high-response features.
• Semantic information and geometric constraints are employed to filter dynamic features.
• An extension method and high-confidence point weighting are used for unmatched features.

Main Results:

• The improved algorithm shows over a 90% performance increase in highly dynamic environments.
• Absolute trajectory error improved by up to 96.84% in low-dynamic settings compared to ORB-SLAM2.
• Demonstrated superior adaptability and robustness in dynamic environments.

Conclusions:

• The proposed dynamic feature detection and filtering algorithm significantly enhances visual SLAM performance.
• The algorithm offers a robust solution for real-world applications involving moving objects.
• This work advances the capabilities of SLAM systems in challenging, non-static scenarios.