The objective of this work is to develop an indoor location system with high precision and continuous position monitoring in real time with the use of a mobile phone without any special hardware using only commercially available low-cost sensors. Finding the location is done using the measured Received Signal Strength Indicator (RSSI) value of Bluetooth beacons received from mobile phones combined with measurements from other phone sensors. For the development of our model, we collected measurements for the RSSI values from Beacons which we placed in a space of 30.75 sqm and the values from the mobile accelerometer in motion. We divided the space into 16 subareas of 1.45m x 1.35m and used our measurements to develop a machine learning model using the open source TensorFlow framework to predict the correct subarea of the user. Through experiments, we show that our model can reach an accuracy of 0.7209 which means that our system can predict the correct user location in 72% of the cases with accuracy less than 1 meter.
Paper
Full text
Bluetooth Beacon Based Accurate Indoor Positioning Using Machine Learning
Semantic Scholar · Computer Science · 2019
Abstract
The objective of this work is to develop an indoor location system with high precision and continuous position monitoring in real time with the use of a mobile phone without any special hardware using only commercially available low-cost sensors. Finding the location is done using the measured Received Signal Strength Indicator (RSSI) value of Bluetooth beacons received from mobile phones combined with measurements from other phone sensors. For the development of our model, we collected measurements for the RSSI values from Beacons which we placed in a space of 30.75 sqm and the values from the mobile accelerometer in motion. We divided the space into 16 subareas of 1.45m x 1.35m and used our measurements to develop a machine learning model using the open source TensorFlow framework to predict the correct subarea of the user. Through experiments, we show that our model can reach an accuracy of 0.7209 which means that our system can predict the correct user location in 72% of the cases with accuracy less than 1 meter.