Location Estimation Applying Machine Learning Using Multiple Items of Sensed Information in Indoor Environments

Abstract

Receiving radio waves directly from satellites is difficult in an indoor environment, making accurate location estimation by satellite signals difficult. Since mobile communication propagation channels suffer from fading and shadowing, estimating accurate locations in an indoor environment by using only the received signal power of a radio wave is also difficult. We have previously proposed a minimum mean square error (MMSE) location estimation method using multiple items of sensed information as well the received signal power and shown that its estimation errors are smaller than those of the conventional method using only the received signal power. Machine learning has recently become attractive as an optimization algorithm, and in this paper we propose to apply machine learning with a back propagation method for indoor location estimation using multiple items of sensed information. Our proposed machine learning location estimation method is experimentally validated and compared with the MMSE method. The machine learning location estimation method was found to reduce the standard deviation of the location estimation error and increase the probability that the estimated distance is within 2.5 m of the actual distance.