GSO-SLAM: Bidirectionally Coupled Gaussian Splatting and Direct Visual Odometry

TL;DR

GSO-SLAM introduces a real-time monocular dense SLAM system that tightly couples Gaussian scene representation with visual odometry using an EM framework, improving accuracy and efficiency.

Contribution

It presents a novel bidirectional coupling of VO and Gaussian Splatting within an EM framework, enabling joint optimization without extra computational costs.

Findings

Operates in real time with high fidelity

Achieves state-of-the-art scene reconstruction quality

Improves tracking accuracy significantly

Abstract

We propose GSO-SLAM, a real-time monocular dense SLAM system that leverages Gaussian scene representation. Unlike existing methods that couple tracking and mapping with a unified scene, incurring computational costs, or loosely integrate them with well-structured tracking frameworks, introducing redundancies, our method bidirectionally couples Visual Odometry (VO) and Gaussian Splatting (GS). Specifically, our approach formulates joint optimization within an Expectation-Maximization (EM) framework, enabling the simultaneous refinement of VO-derived semi-dense depth estimates and the GS representation without additional computational overhead. Moreover, we present Gaussian Splat Initialization, which utilizes image information, keyframe poses, and pixel associations from VO to produce close approximations to the final Gaussian scene, thereby eliminating the need for heuristic methods. Through extensive experiments, we validate the effectiveness of our method, showing that it not only operates in real time but also achieves state-of-the-art geometric/photometric fidelity of the reconstructed scene and tracking accuracy.