• Bluetooth Low Energy (BLE)
• deep learning
• high precision
• Kalman filter
• localization
• regressor

In this thesis, we designed and deployed an indoor high precision localization system for short distances based on RSSI Bluetooth Low Energy (BLE) fingerprinting.
The goal is to create a smart worktable of a factory where the BLE signal is used to detect the position of objects in a short distance to track the production process. The system is composed of a BLE device that emits beacons received by 5 Bluetooth readers. In this work, we trained three regressors, i.e. k-NN, decision tree and random forest, and two different neural networks to determine the position of the beacon. Also given the very noisy nature of RSSI, a Kalman filter was used to reduce the noise. The results obtained offline are promising, the RNN manages to have an error that is less than 0.29m, 90% of the time. Finally, all the techniques were tested in the real-time scenario. This experiment also showed that in real-time, deep learning networks have too high a latency and that the regressors obtain the best performance considering accuracy and time of adaptation to the change of position.