Indoor Positioning and Machine Learning Algorithms
(2021) EITM02 20211Department of Electrical and Information Technology
- Abstract
- This master thesis focuses around improving the efficiency and accuracy of existing indoor positioning systems with the help of Machine Learning (ML). Our work is based on Bluetooth Low Energy (BLE) v5.1. Position estimation is currently being carried out using the Least-Squares (LS) method in the framework. Introducing Machine Learning to position estimation can reduce computation time and increase accuracy of the system because of the additional “learning” in the form of Machine Learning models that is done by the system. An attempt has been made to extract information from the Direction-finding feature that BLE v5.1 presents, and combine it with ML to potentially improve the current position estimation.
In order to test the efficacy of... (More) - This master thesis focuses around improving the efficiency and accuracy of existing indoor positioning systems with the help of Machine Learning (ML). Our work is based on Bluetooth Low Energy (BLE) v5.1. Position estimation is currently being carried out using the Least-Squares (LS) method in the framework. Introducing Machine Learning to position estimation can reduce computation time and increase accuracy of the system because of the additional “learning” in the form of Machine Learning models that is done by the system. An attempt has been made to extract information from the Direction-finding feature that BLE v5.1 presents, and combine it with ML to potentially improve the current position estimation.
In order to test the efficacy of these ML algorithms, a wide range of data has been used for these experiments. This included data from different simulated indoor environments and from measurements done physically in a real office environment. We have experimented with three different Machine Learning algorithms for classification and regression: Random Forest, Support Vector Machine and k Nearest Neighbors. Each algorithm has shown impressive results with centimetre-level accuracy, indicating that it can be rewarding to explore ML even more for the purpose of indoor positioning. (Less) - Popular Abstract
- Indoor positioning comes into play when satellite-based systems such as Global Positioning System (GPS) are not able to provide accurate information. This happens because these systems generally require a direct Line-Of-Sight (LOS) to communicate effectively, or the device to have access to as many satellites as possible. This is not always the case indoors. GPS systems do not work well inside closed structures, especially those built primarily of concrete, because these signals are too weak to be able to penetrate solid structures, which result in high loss in received signal power [1].
In these situations, short-range systems such as Bluetooth and WiFi work very well, where a high level of accuracy cannot be met using GPS. Indoor... (More) - Indoor positioning comes into play when satellite-based systems such as Global Positioning System (GPS) are not able to provide accurate information. This happens because these systems generally require a direct Line-Of-Sight (LOS) to communicate effectively, or the device to have access to as many satellites as possible. This is not always the case indoors. GPS systems do not work well inside closed structures, especially those built primarily of concrete, because these signals are too weak to be able to penetrate solid structures, which result in high loss in received signal power [1].
In these situations, short-range systems such as Bluetooth and WiFi work very well, where a high level of accuracy cannot be met using GPS. Indoor position- ing systems are useful in scenarios where tracking and location-based information of objects are required. This includes Internet of Things (IoT) - e.g. home automation systems - airports, factories, warehouses, places providing healthcare, and transportation facilities [2]. There is a large variety of areas in day-to-day life where positioning using Bluetooth can prove to be helpful, and can make tasks easier. Bluetooth positioning is able to bring the positioning error down to sub-metre accuracy [3]. Indoor positioning with BLE works best when multiple BLE beacons are working together. This usually increases position accuracy. Bluetooth devices carry out positioning using Received Signal Strength Indicator (RSSI), Angle of Arrival or Angle of Departure (AoD).
We have mainly worked on improving the position estimates of a Bluetooth tag provided by four anchor points in an office environment. One of the key features of BLE v5.1 is direction finding, so Bluetooth beacons can provide information about the direction of the incoming signal. We have used the AoAs and RSSI provided by four anchor points to give a better estimate of the position of the Bluetooth device in question using Machine Learning algorithms. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9053855
- author
- Uttarwar, Raavi LU and Valentín, Julián
- supervisor
- organization
- course
- EITM02 20211
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Indoor Positioning, Machine Learning, Bluetooth Low Energy
- report number
- LU/LTH-EIT 2021-822
- language
- English
- id
- 9053855
- date added to LUP
- 2021-06-16 14:05:23
- date last changed
- 2021-06-16 14:05:23
@misc{9053855,
abstract = {{This master thesis focuses around improving the efficiency and accuracy of existing indoor positioning systems with the help of Machine Learning (ML). Our work is based on Bluetooth Low Energy (BLE) v5.1. Position estimation is currently being carried out using the Least-Squares (LS) method in the framework. Introducing Machine Learning to position estimation can reduce computation time and increase accuracy of the system because of the additional “learning” in the form of Machine Learning models that is done by the system. An attempt has been made to extract information from the Direction-finding feature that BLE v5.1 presents, and combine it with ML to potentially improve the current position estimation.
In order to test the efficacy of these ML algorithms, a wide range of data has been used for these experiments. This included data from different simulated indoor environments and from measurements done physically in a real office environment. We have experimented with three different Machine Learning algorithms for classification and regression: Random Forest, Support Vector Machine and k Nearest Neighbors. Each algorithm has shown impressive results with centimetre-level accuracy, indicating that it can be rewarding to explore ML even more for the purpose of indoor positioning.}},
author = {{Uttarwar, Raavi and Valentín, Julián}},
language = {{eng}},
note = {{Student Paper}},
title = {{Indoor Positioning and Machine Learning Algorithms}},
year = {{2021}},
}