RNN-based sensor fusion method to handle inconsistently scanned sensing data for indoor positioning system

Abstract

There have been many attempts to locate mobile devices by using wireless signals indoors. Recently, there are also trials to combine various sensors such as magnetometer and barometer along with wireless LAN signals to further enhance the positioning accuracy. However, sensor fusion for indoor positioning has to overcome problems of deciding weights of wireless signals and sensor values, and handling of different input rates. In this paper, we propose a machine learning method that can integrate wireless signals and sensing data with different scan rates and propose a positioning engine using it. In order to verify the effectiveness of the proposed method, we collected signal data in two types of indoor space characterized by narrow corridor and wide open space respectively and measured the accuracy of positioning using the proposed method. In a comparison with the existing positioning methods, the advantages and disadvantages of this system are analyzed. We also measured the improvement of positioning accuracy by the adding of new sensors.