Enhancing Precision and Robustness in Indoor Localization Using Temporal and Spatial Features for BLE Fingerprinting

In recent years, smartphone indoor localization solutions using radio signals have garnered much interest, with Bluetooth Low Energy proving to be a promising candidate for the task, namely for its low cost, ease of deployment, and wide availability. While many solutions using BLE technology have demonstrated satisfactory localization precision, BLE fingerprint systems are subject to various noise-inducing factors such as signal interference and multipath propagation. In this paper, we propose a solution to present received signal strength fingerprints in order to incorporate spatial information into the input data while using convolutional autoencoders to denoise the data. We evaluate the proposed approach on a dataset collected in a six-story building with an area of approximately 600 square meters. The results obtained show improved performance when preprocessing the fingerprint data so that it incorporates the underlying spatial relationship of the beacons.