Paper Info
Title
An End-to-End BLE Indoor Location Estimation Method Using LSTM
Abstract
Indoor location estimation has long been researched to realize location-based services. In this paper, we propose an indoor location estimation method for Bluetooth Low Energy (BLE) devices using end-to-end LSTM neural network. We focus on large-scale exhibition where is a tough environment for wireless indoor location estimation due to signal strength instability. To achieve higher accuracy, deep learning based methods are proposed rather than trilateration or fingerprint. Existing deep learning based methods estimate the location from the probabilities using the difference of query signal strength and autoencoder-reconstruction of it. Proposed method adopts end-to-end location estimation, which means the neural network takes a time-series of signal strength and outputs the estimated location at the latest time in the input time-series. We also build a loss function which takes how a person walks into account. Considering the difficulty of data collection within a short preparation term of an exhibition, the data generated by a simple simulation is used in the training phase before training with a small amount of real data. As a result, the estimation accuracy is average of 1.92m, using the data collected in GEXPO exhibition in Miraikan, Tokyo. Proposed method outperforms our previous trilateration based method's 4.51m average.
Year
DOI
Venue
2019
10.23919/ICMU48249.2019.9006638
2019 Twelfth International Conference on Mobile Computing and Ubiquitous Network (ICMU)
Keywords
DocType
ISBN
location estimation,BLE,deep learning,LSTM,end-to-end location estimation
Conference
978-1-7281-4226-5
Citations 
PageRank 
References 
0
0.34
11
Authors
4
Name
Order
Citations
PageRank
Kenta Urano184.17
Kei Hiroi21912.00
Takuro Yonezawa38422.34
Nobuo Kawaguchi431364.23