An Indoor Positioning with a Neural Network Model of TensorFlow for Machine Learning

Utilization of machine learning is effective in improving the accuracy of indoor positioning systems. We constructed an experimental positioning trial system in our laboratory for the study of watching over the elderly using a wearable sensor with the air interface of EnOcean wireless standard. The results confirmed that the position estimation accuracy was improved by applying machine learning using a neural network model of TensorFlow. The neural network enables highly accurate position estimation in consideration of the complex indoor radio wave environment by learning the mapping from the RSS data space to the physical space. In this paper, we show the necessity of machine learning based on the observed RSS data set and illustrate the effect of improving accuracy for the case of EnOcean air interface.