The public defense of Hamidur Rahman's doctoral thesis
The public defense of Hamidur Rahman's doctoral thesis in Computer Science and Engineering will take place at Mälardalen University, room Delta (Västerås Campus) at 13.15 on April 7, 2021.
Title of the thesis: “Artificial Intelligence for Non-Contact based Driver Health Monitoring”.
The faculty examiner: Docent Hasan Fleyeh, Dalarna University,
Examining committee: Associate Professor Kerstin Bach, Norwegian University of Science and Technology , Professor Peter Anderberg, University of Skövde; Associate Professor Abdellah Idrissi, Mohammed V University in Rabat; Associate Professor Anne Håkansson, KTH Royal Institute of Technology.
Reserve: Professor Wasif Afzal, Mälardalen University.
Serial number: 330.
Modern vehicles have been equipped with advanced technical features to make them faster, safer, and more comfortable. However, to enhance transport security and traffic safety, i.e., to avoid unexpected traffic accidents it is necessary to consider ‘vehicle driver’ as a part of the environment.
Therefore, it is important to monitor driver’s physical health and mental state and adjust necessary vehicular features or take action e.g. possible to control speed accordingly. To monitor driver’s physical health and mental state, the two approaches commonly used in the literature are: driving behavior-based and physiological parameter-based approach. The physiological parameter-based approach is one of the reliable approaches that give a more accurate indication of physical and mental health. However, for this, sensors are often attached to the human body which is often troublesome and inconvenient and prone to contaminate with artifacts in dynamic situations like in driving situations.
To eliminate such drawbacks, ‘visionbased physiological parameter’ extraction offers a new paradigm for driver’s physical health and mental state monitoring. This thesis report presents an intelligent non-contact-based approach to monitor driver’s cognitive load based on two different types of parameters, i.e., vision-based physiological parameters and vehicular parameters.The contribution of this research study is in four folds:
- implement non-contact based methods using a camera to extract physiological parameters (e.g., heartbeats) in several challenging conditions, i.e., illumination, motion, vibration, and movement;
- implement a non-contact based method for vehicular parameters (e.g., lateral speed, steering wheel reversal rate) extraction from a driving simulator;
- implement a non-contact based feature extraction method based on eye-movement parameters (i.e., saccade and fixation);
- driver’s cognitive load classification using different machine learning and deep learning algorithms.
The results show that the proposed non-contact camera-based approach is 95% accurate compared to the reference wired sensors. Also, in terms of cognitive load classification, the highest achieved average accuracy is 94% using a machine learning algorithm i.e., Logistic Regression considering. On the other hand, using deep learning i.e., Convolutional Neural Networks obtained average accuracy is 91%. In the future, the proposed approach can be evaluated during naturalistic driving conditions considering challenges like high temperature, complete dark/bright environment, unusual movements, facial occlusion by hands, sunglasses, scarf, beard, etc. which the algorithm can today handle in simulated driving situations. As a proof of concept, this research shows that the camera-based non-contact approach using image processing, computer vision, ML, deep learning has a huge potentiality in driver’s physical health and mental state monitoring.