GG-SLAM: Gaussian Grouping과 Splatting을 이용한 객체 수준 SLAM 시스템

Abstract

Simultaneous Localization and Mapping (SLAM) is crucial for applications such as autonomous driving, robot navigation, and 3D reconstruction. With advancements in Virtual Reality (VR) and Augmented Reality (AR), the demand for immersive and realistic experiences is increasing. This paper introduces a real-time SLAM system enhanced with object-level information using Gaussian Grouping, based on the Gaussian Splatting technique. Our system utilizes predefined masks generated by Tracking Anything to segment objects effectively, enabling the real-time construction of high-quality 3D maps. By applying Gaussian Grouping, the system efficiently identifies static objects while integrating object-level data, maintaining the accuracy of traditional Gaussian Splatting-based SLAM. Experimental results demonstrate the system’s high accuracy and efficiency across various environments, enhancing object recognition and interaction for robots, as well as enabling natural interactions with virtual objects in VR/AR applications. This study advances SLAM technologies by providing a high-quality 3D reconstruction method that incorporates object-level information, expanding its applicability to diverse domains.