UTILIZATION OF GAUSSIAN SPLATTING IN VISUAL SLAM

Date

2025-1-7

Author

Sarıkamış, Furkan Aykut

Metadata

Show full item record

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

Item Usage Stats

64
views

65
downloads

Cite This

3D Gaussian Splatting has emerged as a promising alternative to neural implicit representations in SLAM systems. However, current methods often lack dense depth maps or tailored designs for large-scale environments. To address these issues, this thesis introduces IG-SLAM, an RGB-only SLAM system that combines robust tracking methods with Gaussian Splatting. The system builds a 3D map using accurate poses and dense depth from tracking while leveraging depth uncertainty for improved reconstruction. Our decay strategy enhances convergence and enables real-time operation at 10 fps in a single process. IG-SLAM achieves competitive results with state-of-the-art RGB-only SLAM systems at significantly faster speeds, demonstrating photo-realistic 3D reconstruction on several datasets. The project is available at https://github.com/Liouvi/IG-SLAM.

Subject Keywords

Gaussian Splatting, Neural Radiance Fields, Visual SLAM, 3D reconstruction

URI

https://hdl.handle.net/11511/113440

Collections

Graduate School of Natural and Applied Sciences, Thesis