Design and implementation of an object-level indoor navigation algorithm using spherical images

Date

2025-9-01

Author

Kurt, Umut

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

110
views

0
downloads

Cite This

In this thesis, a topological navigation algorithm that leverages both object-level semantic information and pixel-level keypoint features based on spherical images is proposed. A hierarchical map representation is employed for combining these two levels of abstractions. In the lower level of the hierarchy, salient visual features of the scene are extracted and matched across frames using deep-learning models for estimating the relative transformation between robot poses. In the higher level of the hierarchy, object representations are aggregated from multiple observations and utilized for long-term localization and loop closures based on objects that are co-visible by the robot. Moreover, these compact and global object representations have been shown to be useful for target navigation to objects in both simulation and real-world experiments. The map generated with this framework is not limited to spherical cameras and can support the navigation of any robot platform equipped with a camera, regardless of its field of view.

Subject Keywords

Topological Mapping, Spherical Imaging, Object-Level Semantics, Visual SLAM, Hierarchical Map Representation

URI

https://hdl.handle.net/11511/116084

Collections

Graduate School of Natural and Applied Sciences, Thesis