|Abstract:||A novel indoor localization concept is introduced in this study. The navigating user takes an image inside a building to determine its own position. In the first step, this image is passed to a trained R-CNN that recognizes and localizes various types of objects in the image, such as doors, windows, signs, trash bins. Typically, the R-CNN output is a set of boxes representing the regions of the objects in the image, and their classification scores. The spatial relationships of these objects can be derived based on the box locations in the image and they can be represented as a directed graph, where the directions of the edges are the relationships, e.g., right-left, up-down. This graph is called image graph. The next step of the localization requires prior knowledge, basically the building model, obtained by blueprint or surveying, as the location of the used objects must be known. Assuming fixed positions, a graph can be derived similar to the image graph, but it is generated based on the building map. This graph is called position graph. Each unique position inside the building has its own position graph. Finally, the user location is obtained by measuring the similarity between the image graph and the stored position graphs. To prove the concept, a test was conducted in a typical yet challenging building. Images of selected objects were taken at randomly chosen locations around the test area. Correct solution was found in the 40.5% of the cases with 5-10 m accuracy; the failure rate was 14.3%; the rest 45.2% produced multiple solutions. Note that the system is fail-safe at 85.7%. The concept also allows deriving the direction of the image; the results suggest a 45-60º accuracy.|
Proceedings of the 2017 International Technical Meeting of The Institute of Navigation
January 30 - 2, 2017
Hyatt Regency Monterey
|Pages:||1269 - 1279|
|Cite this article:||
Xu, Haowei, Koppanyi, Zoltan, Toth, Charles K., Brzezinska, Dorota, "Indoor Localization using Region-based Convolutional Neural Network," Proceedings of the 2017 International Technical Meeting of The Institute of Navigation, Monterey, California, January 2017, pp. 1269-1279.
ION Members/Non-Members: 1 Download Credit