Visual Indoor Positioning Method Using Image Database

摘要

With the increasing significance and usage of ordinary location-based services, a number of technologies have been invented for indoor positioning systems over the years. Some of them, however, suffer from limited precision and high expenses. In this paper, we propose a simple visual indoor positioning method using images captured by handheld cameras, in which case, there is no need for infrastructure components. Taking advantage of Bag-of-Visual-Words model, our proposed method consists of two stages. By using image processing algorithms i.e. feature extraction, description and clustering, an image database and its visual vocabulary are prepared in the first stage, along with the corresponding indoor position information. Then, a query image is matched with the image database from stage one to get the most several similar ones, whose position will be taken to get the location estimation by the method of voting and re-classification. We conducted some experiments to demonstrate the feasibility of the visual indoor positioning method. According to the simulation results, 90% of the positioning errors are within 1.2 meters and the average positioning time is no more than 2 seconds, which turns out to achieve a balance between speed and accuracy.