🤖 AI Summary
This work addresses the inefficiency and limited accuracy of existing monocular dense SLAM systems, where visual odometry and mapping are loosely coupled. To overcome these limitations, we propose a real-time system that tightly integrates direct visual odometry with Gaussian splatting-based scene representation through a novel bidirectional coupling mechanism. Within an Expectation-Maximization (EM) framework, our approach jointly optimizes depth estimation and reconstruction in an end-to-end manner, without additional computational overhead. Key innovations include the first-ever bidirectional coupling between odometry and mapping, a heuristic-free strategy for initializing Gaussian splats, and a keyframe-based pixel association scheme. The resulting system achieves state-of-the-art performance in tracking accuracy, geometric fidelity, and photometric reconstruction quality while maintaining real-time operation.
📝 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.