An End-to-End BLE Indoor Localization Method Using LSTM

This paper proposes an indoor localization method for Bluetooth Low Energy (BLE) devices using end-to-end LSTM neural network. We focus on a large-scale space where there is tough environment wireless due to signal instability. Our proposed adopts localization, which means input time-series of strength and output the estimated location at latest time in input. The network our consists fully-connected layers. use custom-made loss function with 3 error components: MSE, direction travel, leap location. Considering difficulty data collection short preparation term, generated by simple simulation used training phase, before small amount real data. As result, estimation accuracy achieves average 1.92m, collected GEXPO exhibition Miraikan, Tokyo. also evaluates assuming troubles operation.