Comparison of Different Machine Learning Algorithms for an Efficient Bluetooth Low Energy Based Indoor Positioning System
As GPS fails to navigate inside buildings, indoor localization is the way to realize the goal. It is the use of WiFi, Infrared, or Bluetooth to create digital maps of spatial frames that can be utilized for turn-by-turn navigation indoors. The indoor localization system being developed at NXP will use Bluetooth Low Energy (BLE) for low cost and low power indoor navigation. The system uses Raspberry Pi as signal scanners and NXH3670, an NXP manufactured board for signal advertising. The work highlighted in this Master’s thesis involves the comparison of different Machine Learning (ML) algorithms for the prediction of user location based on RMSE. The goal is to find the most efficient algorithm for the accurate prediction of a target object using RSSI values. Two approaches were presented, supervised, and semi-supervised regression. The results were satisfactory for the first approach with potential prospects in the future, while the second approach was found to be slightly better with a lower RMSE. This work thereby shows that both techniques are capable of being used as an efficient indoor localization system after necessary optimizations.
