The traditional indoor magnetic positioning system cannot give good accuracy in wide space because of the anomaly in geomagnetic data. We propose long short-term memory (LSTM) based deep recurrent neural network (DRNN) model for indoor position system (IPS), which is capable of capturing discriminative features in long-range input sequences. We collect geomagnetic data from various positions inside the building to create a magnetic map and evaluate the performance or our model based on classification accuracy. The proposed model achieved the accuracy of over 90% in a two-dimensional environment.
Click the PDF button to read the full paper.
