Date of Award

5-25-2024

Publication Type

Thesis

Degree Name

M.A.Sc.

Department

Industrial and Manufacturing Systems Engineering

Keywords

Facial Features;Human Driver;Mental Workload;Rocket Classifier;Summarized data;Time Series data

Supervisor

Eunsik Kim

Supervisor

Yong Hoon Kim

Abstract

Safety while driving is of paramount importance. Driver’s mental workload is a vital component in ensuring safe driving. Recent studies implemented machine learning algorithms to make binary classifications of mental workload using feature extracted physiological and video-based measures. Despite achieving decent results, binary classification limits the exploration of optimal workload required for driving. As for the use of feature extracted data, significant information is lost in feature extraction process, which affects the final efficacy of the model developed. Moreover, the use of physiological sensors brings the element of intrusiveness while collecting data in the real-life scenario. The study aims to find the optimal model which accurately classifies mental workload utilizing facial features combined with a machine learning algorithm. Furthermore, the study aims to find the most feasible data type which yields the highest accuracy. A driving simulator study was conducted with four scenarios stimulating various levels of cognitive load measured using a webcam and an EDA sensor. The results showed that time series facial data can classify three levels of mental workload with an accuracy of 86%. The study can serve as a strong basis of employing non-intrusive facial measures to estimate the human driver’s mental workload in the real-world circumstances.

Download

Share

COinS