Transfer Learning for Indoor Localization with Smartphones
Problem Statement:
Indoor localization is essential for many applications. Unlike outdoor localization, indoor localization is a daunting challenge since the Global Positioning System (GPS) signal is attenuated and scattered by roofs and walls. The WiFi radio frequency of smartphones is a popular mean for indoor localization because of its popularity, every smartphone has a WiFi interface. However, WiFi signals are also attenuated and scattered significantly in indoor environments due to walls, furniture, and human bodies. That makes the location estimation sensitive to the chance of the indoor environments. Recently, fingerprinting with deep learning has been used to build a localization model that can learn the environmental parameters for better location estimation.
This project aims at exploiting transformer, an advanced deep learning technique for time-series data, for indoor localization.
Tasks:
The project will be divided into 3 main tasks:
- Study fingerprinting localization with Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM)
- Develop transformer for fingerprinting localisation
- Compare transformer with CNN and LSTM based fingerprinting localisation
WORK:
30% Theory, 50% Implementation, 20%Writing
Contact
Le Viet Duc, v.d.le@utwente.nl, room ZI 5013