We present a semi-automatic method to generate interactive virtual tours from omnidirectional video. Our system is inspired by Google Street View, and its key characteristic is the focus on indoor environment applications: it allows a user to virtually navigate through buildings and indoor scenes. The user can freely look around and navigate to any prerecorded locations. The method automatically computes the initial tour topology from omnidirectional video data using structure from motion. A place recognition step afterwards detects junctions and loop closures. A final interactive refinement step allows one to align the initial topology to a floor plan with a few mouse clicks. This refinement incorporates automatic constraints from the place recognition step in association with manual control-point constraints. The approach presented combines a high degree of automation with a final manual polishing step to create a simple-yet-efficient system that can be applied in various contexts such as virtual tourism, tele-immersion, tele-presence and e-heritage.
ASJC Scopus subject areas