Monocular GSO | Yan Song Hu

The culmination fo my Master’s work.

M-GSO is a real-time volumetric dense simultaneous localization and mapping system. It harness the synergy between the photometric SLAM system Direct Sparse Odometry (DSO) and 3D Gaussian Splatting to achieving a combination of speed, map quality, and memory efficiency unmatched by other dense SLAM systems.

DSO is currently under review for the 2025 IEEE International Conference on Robotics and Automation (ICRA)

Preprint avalaible at MGSO (Hu et al., 2024)

Examples of M-GSO running on the synthetic Replica dataset

Examples of M-GSO running on the real-life EUROC MAV drone dataset

References

2024

MGSO: Monocular Real-time Photometric SLAM with Efficient 3D Gaussian Splatting

Yan Song Hu, Nicolas Abboud, Muhammad Qasim Ali, and 5 more authors

2024

Bib

@misc{hu2024mgsomonocularrealtimephotometric,
  title = {MGSO: Monocular Real-time Photometric SLAM with Efficient 3D Gaussian Splatting},
  author = {Hu, Yan Song and Abboud, Nicolas and Ali, Muhammad Qasim and Yang, Adam Srebrnjak and Elhajj, Imad and Asmar, Daniel and Chen, Yuhao and Zelek, John S.},
  year = {2024},
  eprint = {2409.13055},
  archiveprefix = {arXiv},
  primaryclass = {cs.RO},
  url = {https://arxiv.org/abs/2409.13055},
}