A Multi-Strategy Visual SLAM System for Motion Blur Handling in Indoor Dynamic Environments

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) systems typically assume static environments, limiting their use in dynamic indoor settings common for household robots.
• Existing methods use semantic information to handle dynamic objects but struggle with motion blur, which hinders reliable semantic extraction.
• Accurate environmental perception is crucial for autonomous navigation and scene understanding in domestic robots.

Purpose of the Study:

• To propose a novel visual SLAM algorithm capable of robustly handling dynamic indoor environments.
• To improve the reliability of semantic information extraction in the presence of motion blur.
• To reduce the impact of motion blur on SLAM system performance for enhanced robot autonomy.

Main Methods:

• Integration of a missed segmentation compensation mechanism for predicting and restoring semantic information.
• Leveraging depth and semantic data to generate dynamic object masks.
• Incorporation of a probability-based algorithm for dynamic feature detection and elimination to refine keypoint filtering.

Main Results:

• The proposed SLAM system demonstrated lower Absolute Trajectory Error (ATE) compared to existing systems in dynamic indoor environments.
• The system showed superior performance, especially in scenarios with significant view angle variations.
• Evaluation on TUM and Bonn RGB-D datasets validated the effectiveness of the proposed methods.

Conclusions:

• The novel visual SLAM algorithm effectively addresses challenges posed by dynamic environments and motion blur.
• The system enhances semantic information reliability, crucial for accurate mapping and localization.
• The proposed approach can significantly improve the autonomous navigation and scene understanding capabilities of domestic robots.