Late Breaking Results on Visual S-Graphs for Robust Semantic Scene Understanding and Hierarchical Representation

The current work introduces an improved version of our previous marker-based approach presented in IROS'23 with a broader coverage of sensors and semantic concepts support built upon ORB-SLAM 3.0. It employs pose information of the markers placed on walls and doorways to estimate the 3D equation of their planes, while the markers only keep their correspondence information of the semantic entities. Compared to its baseline and other similar works, the proposed approach produces higher-accuracy reconstructed maps with semantic entities, including walls, rooms, corridors, and doorways.