An Efficient Indoor Localization Method Based on the Long Short-Term Memory Recurrent Neuron Network

With the development of deep learning, fingerprints recognition based on neural networks is a widely used method in indoor localization. In this paper, we build a long short-term memory (LSTM) recurrent neuron network to make regression between fingerprints and locations in order to track the moving target. Simulations are in a BLE5.0 based environment and we use received signal strength indication (RSSI) as the element of fingerprints. Since the preparation of fingerprints is an inevitable and time-consuming process in the testing phase of LSTM, we propose two methods to improve the real-time performance of the localization without changing the structure of LSTM. A decentralized sorting algorithm is proposed to divide the received RSSI signals into multiple parts based on the MAC address of BLE5.0 equipment and use GPUs to sort each part. A complete fingerprint is a combination of these parts. Then, an optimization model aimed at maximum localization accuracy and minimal time used in the testing process of LSTM is proposed by changing the length of fingerprints. Many experiments simulated in different trajectories show that LSTM is more accurate in localization than many other neural networks. Further results demonstrate that using decentralized fingerprints preparation and finding an optimal fingerprint length can keep balance between the localization accuracy and real-time performance.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.