Improving ORB-SLAM3 Accuracy in Dynamic Scenes with YOLO11 Segmentation

Area of Science:

• Robotics and Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Traditional Visual SLAM systems struggle with accuracy in dynamic environments.
• ORB-SLAM3, a popular SLAM system, is susceptible to performance degradation due to moving objects.

Purpose of the Study:

• To enhance ORB-SLAM3 for robust performance in dynamic environments.
• To integrate instance segmentation for dynamic feature exclusion in SLAM.

Main Methods:

• Developed YOLOv11-ORB-SLAM3 by integrating a YOLOv11 instance segmentation module into ORB-SLAM3.
• Tested the system with stereo and RGB-D cameras on the TUM RGB-D dataset and real-world mobile robot experiments.

Main Results:

• YOLOv11-ORB-SLAM3 significantly outperforms the original ORB-SLAM3 in dynamic scenarios.
• Achieved a 93% error reduction on the TUM RGB-D dataset.
• Maintained computational efficiency.

Conclusions:

• The proposed YOLOv11-ORB-SLAM3 system offers a robust and efficient solution for visual SLAM in dynamic environments.
• Demonstrated viability for real-world robotic applications requiring accurate localization amidst moving objects.