Modeling, Designing and Applying Machine Learning Algorithms for Driver Drowsiness Detection

<p>Driver drowsiness has been a significant hazard resulting in various traffic accidents. Therefore, monitoring this condition is crucial not only in alerting drivers, but also in avoiding fatal accidents. Many research studies propose new systems to reduce the number of drowsiness-related injuries and fatalities. The ultimate goal for a drowsiness detection system is to detect the drowsiness on time and minimize the system or environment errors to avoid false readings, such as studying physiological signal processing patterns. These potentially life-saving systems must operate in a timely manner with the highest precision. Researchers proposed various methods based on driving pattern changes, driver body position, and physiological signal processing patterns. There is a focus on human physiological signals, specifically the electrical signals from the heart and brain. In this study, we are presenting an alternative method to determine and quantify the driver drowsiness levels using a physiological signal that was collected in a non-intrusive method. This methodology utilizes heart rate variation (HRV), electrocardiogram (ECG), and machine learning for drowsiness detection. It is apparent that a driver’s drowsiness is associated with an immediate change in heart rate, and due to the fact that Electrocardiogram (ECG) is used to detect an accurate heart rate. We used it as a parameter in the proposed design where it consists of a non-contact ECG sensor as an input source and a circuit with a two-stage amplifier to improve the ECG signal’s strength and filters to minimize noise. An approximate maximum peak ECG output voltage of 2.8V was obtained in LT Spice, and the resulting ECG output is sufficient enough to detect a driver’s drowsiness while preventing major accidents. Furthermore, the HRV is measured with an ECG. The algorithm uses both wavelet and short Fourier transform (STFT). The algorithm extracts and selects the desired features. Then, the system applies both the support vector machine (SVM) and K- nearest neighbor (KNN) method. This achieves an accuracy of 80% or higher. In this research, the accuracy output for the SVM method is 83.8%, 82.5% when using STFT, and 87.5% when applying the WT technique. The algorithm with highest accuracy helps to decrease the number of accidents due to drowsiness. Furthermore, we applied unsupervised machine learning (clustering) to study the behavior of HRV during drowsiness. We can measure different levels of drowsiness based on the changes in the density and shape of the HRV clusters by using this method. Moreover, the pre-measured labeled data is not required to establish the algorithm in this method. Therefore, this algorithm evaluates drowsiness and no prerecorded data is required for any unknown object or person. Successful application of this drowsiness detection method may help to avoid traffic accidents. This study may be beneficial for policy maker’s in preparing regulations to prevent traffic accidents worldwide and may also helpful for users to increase their knowledge and awareness regarding drowsiness detection. Keywords: Drowsiness, Machine Learning, Electrocardiogram (ECG), Heart Rate Variability (HRV), Wavelet Transform (WT).</p>

Name thesis:degree_name

Engineering and Industrial Applied Mathematics Joint PhD with California State University Long Beach, PhD

Level thesis:degree_level

Open Access Dissertation

Discipline thesis:degree_discipline

Institute of Mathematical Sciences

Year dc:date.available

2020

Author dc:creator

• Babaeian, Mohsen

Contributors dc:contributor

• Allon Percus
• Ali Nadim
• Henry Yeh

• Drowsiness
• Electrocardiogram (ECG)
• Heart Rate Variability (HRV)
• Machine Learning
• Wavelet Transform (WT)
• Biomedical Engineering and Bioengineering

Repository record dc:identifier

https://scholarship.claremont.edu/cgu_etd/642

OAI identifier oai:identifier

oai:scholarship.claremont.edu:cgu_etd-1664