Indoor Image Geocoding using Synthetic Views

Abstract

The Global Positioning System is well known for not reaching indoor environments. Several Indoor Positioning System's have been proposed, but most of these solutions either have high accuracy errors or use expensive material to attenuate positioning errors. In this paper we propose a Computer Vision routine which is able to compute the location and orientation on indoor environments. This routine is based on Structure from Motion, an incremental algorithm which recovers the 3D structure from related photographs. The 3D structures generated are geocoded, stored in a database, and new photographs can be added at any time. By combining these 3D structures with the already existing Synthetic Views method for fast location recognition, we are able to compute the indoor GPS coordinates and orientation of new photographs in less than a second.