SKYLINE2GPS: Localization in urban canyons using omni-skylines

摘要: This paper investigates the problem of geo-localization in GPS challenged urban canyons using only skylines. Our proposed solution takes a sequence upward facing omnidirectional images and coarse 3D models cities to compute geo-trajectory. The camera is oriented upwards capture immediate skyline, which generally unique serves as fingerprint for specific location city. goal estimate global position by matching skylines extracted from omni-directional skyline segments city models. Under day-time clear sky conditions, we propose sky-segmentation algorithm graph cuts estimating geo-location. In cases where gets affected partial fog, night-time occlusions trees, shortest path that computes without prior detection. We show compelling experimental results hundreds taken New York, Boston Tokyo under various weather lighting conditions (daytime, foggy dawn night-time).