Qualitative Image-Based Localization in a Large Building

摘要

Interest in indoor localization is growing because it is an important component of many applications. Image-based localization, using naturally-occurring features in the environment, is an attractive solution to this problem. A challenge is to be able perform this on a mobile device with limited computing power. Another challenge is that buildings can have locations with a similar appearance, which can confuse an image-based recognition system. Since many applications do not need exact location, we focus on qualitative localization, which is the problem of the problem of determining approximate location by matching a query image to a database of images. We propose a novel approach that uses an efficient hashing scheme to quickly identify candidate locations, then applies a strong geometric constraint to reject matches that have similar appearance. On experiments in a large campus building, we show that this approach can localize a query image with high accuracy and has potential to run in real time on a mobile device.