MG-SLAM: Structure Gaussian Splatting SLAM With Manhattan World Hypothesis

Gaussian Splatting SLAMs have made significant advancements in improving the efficiency and fidelity of real-time reconstructions. However, these systems often encounter incomplete reconstructions in complex indoor environments, characterized by substantial holes due to unobserved geometry caused by obstacles or limited view angles. To address this challenge, we present Manhattan Gaussian SLAM, an RGB-D system that leverages the Manhattan World hypothesis to enhance geometric accuracy and completeness. By seamlessly integrating fused line segments derived from structured scenes, our method ensures robust tracking in textureless indoor areas. Moreover, The extracted lines and planar surface assumption allow strategic interpolation of new Gaussians in regions of missing geometry, enabling efficient scene completion. Extensive experiments conducted on both synthetic and real-world scenes demonstrate that these advancements enable our method to achieve state-of-the-art performance, marking a substantial improvement in the capabilities of Gaussian SLAM systems. Note to Practitioners—This paper was motivated by the limitations of Gaussian Splatting SLAM systems in complex indoor environments, where textureless surfaces and obstructed views often lead to substantial tracking errors and incomplete maps. While existing systems excel in high-fidelity reconstruction, they struggle with frame-to-frame or point-feature tracking in large-scale environments, particularly with significant camera rotations and obscured structures. In this paper, we enhance the neural dense SLAM by integrating the Manhattan World hypothesis, applying its parallel line and planar surface constraints for more robust tracking and mapping. We incorporate line segment features into both tracking and mapping to improve structural accuracy. Moreover, we propose a post-optimization method that interpolates new Gaussian primitives to effectively fill gaps on planar surfaces. Extensive experiments on multiple datasets demonstrate the superiority of our approach in large-scale indoor environments, resulting in more accurate tracking and mapping.