단일 채널 뇌파 신호를 활용한 졸음 운전 감지를 위한 특징점 기반 필터링 알고리즘 타당성 연구

Abstract

Drowsy driving is one of the major causes of traffic accidents. And a lot of drowsy driving accidents occur during high-speed driving, and it is easy to get fatal injured. For that reason, it is important to detect drowsiness quickly during driving. However, the previous studies required more than 20 seconds of EEG signals to detect drowsiness during driving. In this study, we conducted real driving experiments on 18 subjects. And the aims of this thesis are to, 1) confirm the difference of EEG features between awake driving condition and drowsy driving condition and 2) propose a novel method, common feature pattern (CFeP) filter, to detect driver's drowsiness during driving by using only single electrode EEG signal. To achieve the aims, first, 26 features were extracted in order to identify the difference of EEG features between awake driving condition and drowsy driving condition. In here, 26 features were made by the combination of 5 frequency bands(delta band (1~4 Hz), theta band (4~8 Hz), alpha band (8~12 Hz), beta band (12~30 Hz), and gamma band (30~50 Hz)) and 5 features (frequency band power ratio (BPR), dominant frequency (DF), average power of dominant peak (APDP), center of gravity frequency (CGF), and frequency variability (FV)), and alpha/beta BPR. Also using the extracted features, I verified the change of feature values and the significance of each feature. Second, I proposed a novel method, common feature pattern (CFeP) filter, to detect driver’s drowsiness during driving by using single electrode EEG signal. The CFeP filter is using 26 features - time matrix extracted from a single channel electrode EEG signal. And I verified the accuracy by changing the EEG signal length used in the algorithm from 3 seconds to 10 seconds. As a result, 26 features showed a significant difference between awake driving condition and drowsy driving condition. In the accuracy, when using 3 seconds EEG signal, the accuracy of C5 channel electrode on left central area was 87.34% and when using the EEG signal of 10 seconds, the accuracy of the C1 channel electrode on left central area was 95.35%. The results of this thesis indicate that all of 26 features are important to detect drowsiness during driving. Also, the position of EEG channel electrode may affect the drowsiness detect accuracy, when the drowsiness of the driver is detected using the single channel during operation. Lastly, the novel CFeP filter, it takes less time to detect drowsiness than previous drowsiness detect algorithms. And the drowsiness detect accuracy is also higher than previous drowsiness detect algorithms. In summary, the CFeP filter is considered to be a useful algorithm for drowsiness detect than the previously proposed algorithm.