Unit:
Κατεύθυνση / ειδίκευση Τεχνολογίες Πληροφορικής και Επικοινωνιών (ΤΠΕ)Πληροφορική
Author:
Kiomourtzis Georgios
Kouros Leonidas
Supervisors info:
Αλωνιστιώτη Αθανασία
Original Title:
Ανάπτυξη συστήματος αποφυγής σύγκρουσης στο χώρο μέσω ανάλυσης Δεδομένων Ηλεκτροεγκεφαλογράφου με χρήση Μηχανισμών Μηχανικής Μάθησης
Translated title:
Collision Avoidance System implementation based on EEG frequency Analysis using ML Techniques
Summary:
In this thesis we describe the implementation of a collision avoidance system using
electroencephalography (EEG) signals. EEG signals are gathered from the driver of a
vehicle and transmitted through a mobile application service running in the background
of the driver’s mobile device.
Assuming the EEG signals of the driver’s brain activity can be accessed by a mobile
device, we use that data to classify the driver’s eye state (open or closed). Our aim is to
alert the targeted driver as well as neighbouring drivers of an eye closed state and a
vehicle control loss probability.
The drivers’ mobile devices have a bi-directional communication with an edge server,
responsible for collecting the EEG data, and sending the appropriate alert signal to the
devices after classifying the eye state of the drivers. In order to achieve that, the server
uses different Machine Learning models, to which it is already trained with an
appropriate dataset provided for the purposes of analogous projects.
Collected data and eye state classifications are also send to a Backhaul Server,
responsible for storing all data in database entries and providing the Edge Server with
the best Machine Learning model for data classification.
We present the design and the implementation of the architecture of the system and its
composites, including the client-side mobile device installed application, the signal
analysis and processing, as well as the generation of the best classification model by
training some of the most well-known ML algorithms.
We managed to achieve a successful real-time communication between the mobile
device and the server measuring a metric-defined score of 0.795 accuracy prediction
with the use of Support Vector Machines classifier algorithm.
This undoubtedly allows a margin for improvement, shows however the potential of
developing a reliable Collision Avoidance System through the use of advanced Machine
Learning Classification models combined with the provision of enlarged sizes of EEG
data for the ML models to be trained on.
Main subject category:
Technology - Computer science
Keywords:
Internet of Things, signal processing, principal component analysis, machine learning
eeg_v7.pdf (779 KB)
Open in new window
eeg_src.zip
2 MB
File access is restricted.
