GSOMAR: a geospatial SLAM framework for real-time outdoor mobile augmented reality scene modeling and semantic interaction

This study proposes Geospatial Simultaneous Localization and Mapping for Outdoor Mobile Augmented Reality (GSOMAR), a framework that integrates geospatial data with perceptual computing to enable real-time modeling and interaction in complex urban environments. Unlike traditional mobile augmented reality (MAR) systems based on isolated overlays, GSOMAR ensures spatial continuity and semantic integration of heterogeneous elements – such as buildings, roads, points of interest (POIs), and pedestrians – within a coherent and responsive MAR environment. The framework fuses real-time kinematic global navigation satellite system (RTK-GNSS) measurements with an inertial measurement unit (IMU) to achieve accurate pose estimation, calibrates global scale and orientation, and supports real-time semantic modeling for interactive MAR scenes. Experimental results show that GSOMAR consistently outperforms existing RTK-GNSS + visual – inertial odometry (VIO) and ORB-SLAM-based methods across diverse outdoor scenarios. It achieves high-precision initialization over varied terrain, reducing average positional and gravity-vector errors by up to threefold compared with baseline methods. Extended tracking tests over 1400 m confirm sub-decimeter accuracy and improved stability with reduced drift. virtual-real alignment experiments further demonstrate robust spatial registration under occlusion and extended-range conditions. Overall, GSOMAR enables accurate, interactive, and semantically aware outdoor MAR, supporting applications such as urban modeling, navigation, and infrastructure inspection.