Machine-Learning Indoor Localisation Using Wi-Fi RSSI Fingerprints

GPS does not work indoors, but every phone hears the Wi-Fi access points around it. This project finds the building, the floor and the (x, y) position of a phone from the received signal strength (RSSI) of 520 access points on a real multi-building university campus (UJIIndoorLoc). The classical ECE method - weighted k-nearest-neighbour (WkNN) fingerprinting - is tuned honestly (RSSI representation, distance, k) and compared with a Random Forest, a multi-task deep neural network and a stacked-autoencoder network on a test set recorded four months later with other phones. The robustness to access points that are switched off is measured, a second database (Tampere) checks that the methods generalise, GNU Octave re-runs the fingerprinting, netsh records real Wi-Fi scans on the laptop, and a web application localises uploaded scans on a floor map. A complete MATLAB version is included.

Category: Electronics and Communication Engineering (Wireless Networks, Wi-Fi RSSI Fingerprinting, Indoor Positioning, IoT and Location-Based Services, Machine Learning, Deep Learning, GNU Octave, Windows netsh Wi-Fi Survey, MATLAB, Python)

Electronics & Communication Engineering Project E04 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.