Fingerprint-based indoor positioning and intensity classification using an improved machine learning framework

Abstract

Covid-19 has changed the world in terms of business, public, and many other fields. Millions of livelihoods have been affected by the pandemic. Amidst the upheaval, work from home and restrictions on indoor gatherings have played a significant role in flattening the curve. Even after enforcing restrictions, numbers are still on the rise. Various covid-19 tracing applications have been designed to keep track of positive cases. There is an increased need of tracking positive covid cases to limit the spread of the virus to ordinary people. The continents are trying to flatten the curve and maintain a good economic condition to attain normalcy in the season of chaos. Technology has proved helpful in times of pandemics. Now we have IoT devices and advanced tech, including cameras, Wi-Fi, Bluetooth, RFID etc., which can be used for tracking positive patients. This tracking should be made efficient without exploiting the privacy of users. Vaccination research along with proper tracking seems to be a failsafe solution for evading covid- 19 after effects. Amidst all these available strategies, Indoor localization seems to be one of the required fields of research. This thesis dives into establishing a machine learning framework that can be used across all kinds of IoT (Internet of Things) systems and WSNs (Wireless sensor networks). Distance estimation based on fingerprint has been a widely researched field for indoor localization algorithms. Several traditional approaches have been tried out , including trilateration, triangulation which needs more testing parameters and render them complex. Fingerprinting techniques seems to be helpful. Even though various fingerprinting techniques have been tried out, we do not have a generic framework that can be used for research on fingerprinting. The system has been implemented as a part of cloud remote monitoring solutions and an accurate blend of ensemble bagging and boosting methods for making an accurate distance estimation based on the strength of RSSI fingerprints. The framework hopes to serve as a base platform for all kinds of indoor localization research. It encompasses a BLE-based system that acquires data from leak detection systems, relays it to the cloud via BLE gateway, accumulates data on a cloud database, and is passed as alert notifications to users via the use of the cloud-designed app. At the other end, the database aids in the creation of a location dataset for machine learning which is used for training the model. A regression machine learning model is deployed for the prediction of distances based on fingerprint strength which can be utilized for various fingerprint algorithms. A classification machine learning model is deployed for fingerprint intensity classification to evaluate fingerprint levels in different environments.