Bridging the Gap: Enhancing Visual Indoor Mapping through Semantic Association and Reference Alignment

•(1) An indoor visual point cloud mapping method was enhanced by semantic and consistent referencing.•(2) Light reflections and missing matches are mitigated by the association between semantics and structure.•(3) Indoor-outdoor consistent mapping based on a dense sequential trajectory. In global navigation satellite system-denied environments, indoor point-cloud maps are valuable sources of detailed and inaccessible information. Dense visual point clouds can be produced quickly and inexpensively using stereo RGB cameras and simultaneous localization and mapping (SLAM) techniques. However, visual point-cloud mapping is affected by complex visual observation conditions, and a lack of a consistent reference system sets limits to its application. To address these challenges, this study proposes a method that integrates semantic associating associations and reference bridging. This method employs associations between locations, structures, and semantics to mitigate the detrimental effects of light reflections and subpar textures on visual point-cloud reconstruction. Furthermore, a bridge was established to align the original visual point-cloud map with the geographic reference, where the alignment was made based on the bridge’s control points and its trajectory. Experimental results showed that the bridged visual point cloud was consistent with the previous geo-referenced map with the absolute transformation error of 0.53cm, and the optimized visual point cloud achieved over 90% in semantic completeness. This approach demonstrates that visual mapping can be more reliable and useful in indoor environments.