Enhancing ORB-SLAM3 with YOLO-based Semantic Segmentation in Robotic Navigation

Abstract

This research paper presents an enhanced version of ORB-SLAM3 by integrating it with YOLOv8 for real-time pose estimation and semantic segmentation. ORB-SLAM3 is a state-of-the-art monocular visual SLAM system that uses an ORB feature detector and descriptor to track the camera motion and estimate the 3D map of the environment. However, ORB-SLAM3 lags with the ability to detect and segment objects, which limits its potential applications in robotics and autonomous systems. To address this limitation, we propose to integrate YOLOv8, a high-performance object detection and segmentation framework, into ORB-SLAM3. Our enhanced system combines the strengths of ORB-SLAM3 and YOLOv8, providing accurate and real-time detection and segmentation of objects in complex environments. We evaluate the system on a publicly available dataset and demonstrate that the enhanced system outperforms the baseline ORB-SLAM3 in terms of accuracy and computational efficiency. The system has potential applications in various fields, including robotics, autonomous systems, and augmented reality.