Indoor Robot Navigation System with Relocalization Based on Semantic-Weighted Pose Estimation Using ORB-SLAM3 and YOLOv8

Setiawan, Yahya (2026) Indoor Robot Navigation System with Relocalization Based on Semantic-Weighted Pose Estimation Using ORB-SLAM3 and YOLOv8. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221112-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Indoor navigation for autonomous mobile robots requires a localization system that remains reliable when standard tracking is lost due to motion blur, occlusion, or rapid camera movement. Visual SLAM frameworks address this through relocalization, but conventional pipelines treat all detected feature points equally, regardless of whether they originate from stable structures or unreliable, dynamic regions. This thesis proposes a semantic-weighted pose estimation method that incorporates YOLOv8 object detection into the relocalization pipeline of ORB-SLAM3, replacing its native procedure. The pipeline combines Bag-of-Words candidate retrieval, a Hamming distance pre-filter, RANSAC-based geometric verification with a Perspective-3-Point solver, and a custom Levenberg-Marquardt optimizer on the SE(3) Lie group using Sophus, where each keypoint is weighted according to whether it falls within a detected object bounding box. Evaluated on the TUM RGB-D benchmark, the proposed pipeline reduces Absolute Trajectory Error RMSE by 98.0% compared to native ORB-SLAM3 relocalization and by 28.8% compared to an unweighted variant, with an optimal background weight of 0.3. The relocalized pose is integrated with an A* path planner on an occupancy grid to form a complete indoor navigation system under ROS2 Humble. These results show that lightweight semantic cues from object detection substantially improve the robustness and accuracy of camera relocalization in indoor environments.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Visual SLAM, Relocalization, Semantic Weighting, Pose Estimation, YOLOv8, ORB-SLAM3, Indoor Navigation.
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Yahya Setiawan
Date Deposited:	30 Jul 2026 01:30
Last Modified:	30 Jul 2026 01:30
URI:	http://repository.its.ac.id/id/eprint/140042

Actions (login required)

View Item