DOE-SLAM: Dynamic Object Enhanced Visual SLAM.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Monocular visual simultaneous localization and mapping (vSLAM) systems struggle in dynamic environments due to moving objects.
• Existing vSLAM methods often fail when background features are obscured by dynamic elements.

Purpose of the Study:

• To develop a novel vSLAM strategy capable of robustly handling dynamic environments.
• To improve camera pose estimation accuracy and robustness in the presence of moving objects.

Main Methods:

• A new vSLAM strategy is formulated to adapt to dynamic environments.
• The system tracks camera trajectory using static background features and estimates foreground object motion.
• When background features are insufficient, it utilizes object features and motion prediction for pose estimation.

Main Results:

• The proposed system demonstrates enhanced pose estimation accuracy and robustness.
• Experiments on synthetic, real-world, and TUM benchmark datasets validate the system's performance.
• Outperforms state-of-the-art monocular vSLAM systems in dynamic scenarios.

Conclusions:

• The novel vSLAM strategy effectively addresses challenges posed by dynamic environments.
• The system provides a more accurate and robust solution for camera localization and mapping.
• This approach significantly advances the capabilities of monocular vSLAM in real-world applications.