Smartphone based Indoor Localization Technology using 1D CNN -BLSTM

Abstract

The study of indoor localization technology using smart phone has been continuously studied. Fingerprinting is a representative indoor positioning technology. This technology estimates the location by comparing Radio Signal Strength (RSS) information received in one-shot at a specific location with the previously constructed Radio Map. Since the RSS received in one-shot is used, the ability to discriminate signals according to space is low. To solve this problem, the use of RSS spatial patterns based on Pedestrian Dead Reckoning (PDR) improves signal discrimination according to space and increases accuracy. However, since PDR is used, there is a problem that it is difficult to use a spatial pattern if PDR distortion occurs due to a heading drift error and a change motion. We propose an indoor positioning technology using 1D Convolutional Neural Network (CNN) and Bi-directional Long Short Term Memory (BLSTM). We estimated the position by learning the 1D RSS pattern. In order to generate a large amount of data, we used the pre-built Radio Map. We use a model that combines 1D CNN and BLSTM. 1D CNN is used to extract RSS patterns, and BLSTM is used to learn the relationship of sequential data in both directions. Through this, it is possible to estimate the position using only the RSS. To verify the proposed technology, we compared it with the previous technology. As a result, the previous technology showed 2.19m error and the proposed technology showed 4.663m error. However, the calculation speed is 30 times faster than the proposed technology. It was confirmed that indoor positioning technology using deep learning technology can provide position information with only 1D RSS pattern.