SmartGuide: Fast Single-image-based Building Localization with Smartphone

We implement a light-weighted and fast approach, SmartGuide, to localize and recognise a distant unknown building, using only one photo of a target building and a local 2D Google map without other prior knowledge. SmartGuide first extracts the top view contours of a building from its sideview photo, by applying vanishing point and Manhattan World Assumption. It then fetches a candidate building set from local 2D Google map based on GPS readings, and extracts their contours. Top view shapes, orientation and distance relative to camera are parameters for probability model, based on kernel density estimation, to eliminate smartphone sensors’ noises in GPS readings and camera ray direction calculated by accelerometer and compass, thus accurately recognize the most possible target building in the local map. Experiment results demonstrate that our approach recognizes buildings ranging from 20m to 520m and achieves 92.7% accuracy. In addition, the processing time is no more than 6 seconds for 87% cases. Compared to other image based object detection and localization schemes, SmartGuide offers numerous advantages, while preserving or improving high accuracy. Our method requires only one photo to do localization, and avoids intricate 3D reconstruction and any initial deployment cost of database construction, which makes it faster than existing solutions.