Research On Fast Single-image-based Building Localization With A Smartphone

Building localization closely relates with people’s daily lives, which is an important research issue. In the information and networking age, the rapid development of Internet and smartphones make it increasingly possible for a user to localize an unknown distant building via its photos. Thus, building localization has received extensive research during recent years.Several approaches to this problem have been proposed. One kind of solution is to search the target building among a prepared image database leveraging computer vision technology. Tons of tags and street view photos on the internet facilitate this solution,however, preparing an indexed mage database and searching among billions of images can be troublesome. The other solutions are to take several photos of the target building and calculate physical depth of the building from camera, then estimate building location based on the user’s position, which is cumbersome and require intricate computations.Therefore, it is challenging to find a fast light-weighted building localization approach.To overcome aforementioned challenges, we introduce Smart Guide, a fast lightweighted approach to localize a distant unknown building. Our approach relies on shooting only a single photo of a target building via a smartphone and online map services.Our method avoids taking multiple photos and any initial deployment cost of database construction, making it faster and less labor-intensive than existing solutions. The main contributions lie as followings:? We researched on extracting building’s top view contour from a single image.Based on pinhole camera model and the Manhattan World Assumption, we extracted the visible top view contour from one single image of the target building.? We designed an approach to localize the target building based on features of the building’s top view. We analyzed top view features(shape, orientation and distance) of the target building and candidate buildings to localize the target building among candidate building set.? We introduced the kernel density estimation model to deal with the smartphone’s sensor data noises.? We implemented and evaluated the entire system on di?erent Android phones. Experimental results demonstrate that our approach recognizes buildings ranging from 20m to 520 m and achieves 92.7% accuracy in downtown areas where the Manhattan World Assumption is applicable. In addition, the processing time is no more than 6 seconds for 87% of cases.