SAVMap: Structure-Aided Visual Mapping of Large-Scale 2.5D Manhattan Wireframes from Panoramic Video
Researchers have developed SAVMap, a novel method for creating detailed 3D wireframe maps of large-scale industrial environments using only panoramic video. This system extracts semantic features from video sequences, such as shelf corners and light centers, and uses these points to reconstruct a 3D wireframe map. SAVMap accounts for real-world geometric constraints like Manhattan grids to improve accuracy. The method was successfully demonstrated in a warehouse setting, mapping over 5000 shelf elements with an average error of just 4.8 cm. AI
IMPACT Enables more precise robotic navigation and digital twin generation in industrial settings.