Pembuatan Dataset Primer dan Klasifikasi Kantuk Berdasarkan Eye Aspect Ratio (EAR) Dan Mouth Aspect Ratio (MAR) Menggunakan Deep Learning

Dewi, Tabitha Natasya Cyntia (2026) Pembuatan Dataset Primer dan Klasifikasi Kantuk Berdasarkan Eye Aspect Ratio (EAR) Dan Mouth Aspect Ratio (MAR) Menggunakan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024221008-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (41MB) | Request a copy

Abstract

Penelitian ini membangun dataset primer kantuk berbasis wajah masyarakat Indonesia serta mengembangkan sistem klasifikasi tingkat kantuk berdasarkan kombinasi Eye Aspect Ratio (EAR) dan Mouth Aspect Ratio (MAR) menggunakan tiga arsitektur deep learning, yaitu Time Series Transformer (TST), CNN-BiLSTM-Attention, dan Temporal Convolutional Network (TCN). Pada penelitian ini, sinyal EAR direpresentasikan sebagai pola biner kedipan yang dipisahkan berdasarkan durasi menjadi empat fitur statistik, termasuk PERCLOS, sedangkan sinyal MAR direpresentasikan menggunakan tiga fitur statistik kontinu. Evaluasi dilakukan melalui dua skenario pengujian, yaitu perbandingan kombinasi fitur dan perbandingan performa arsitektur, menggunakan dataset yang terdiri atas 81 video dari 40 partisipan yang dikumpulkan melalui simulator mengemudi BeamNG.drive. Hasil pengujian menunjukkan bahwa kombinasi fitur EAR+MAR memberikan performa terbaik dengan akurasi sebesar 98,29% dan macro F1-score sebesar 98,02%. Pada perbandingan arsitektur, model TST memperoleh performa tertinggi dengan akurasi 98,29%, diikuti oleh CNN-BiLSTM-Attention sebesar 98,06% dan TCN sebesar 97,63%. Selain itu, model yang diusulkan juga melampaui performa baseline pada dataset sekunder UTA-RLDD yang mencapai akurasi 65,20%. Hasil tersebut menunjukkan bahwa penelitian ini berhasil membangun dataset primer yang representatif, mengidentifikasi kombinasi fitur yang optimal, serta mengevaluasi dan membandingkan performa tiga arsitektur deep learning untuk klasifikasi tingkat kantuk pengemudi.
==================================================================================================================================
This study constructs a primary driver drowsiness dataset based on Indonesian facial characteristics and develops a drowsiness classification system using a combination of Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) features with three deep learning architectures: Time Series Transformer (TST), CNN-BiLSTM-Attention, and Temporal Convolutional Network (TCN). In the proposed approach, the EAR signal is represented as a binary blink pattern and transformed into four statistical features, including PERCLOS, based on blink duration, while the MAR signal is summarized into three continuous statistical features. The proposed method is evaluated through two experimental scenarios: feature combination comparison and deep learning architecture comparison, using a primary dataset consisting of 81 videos collected from 40 participants in a BeamNG.drive driving simulator. The experimental results demonstrate that the combined EAR+MAR features achieve the best performance, yielding an accuracy of 98.29% and a macro F1-score of 98.02%. Among the evaluated architectures, TST achieves the highest accuracy (98.29%), followed by CNN-BiLSTM-Attention (98.06%) and TCN (97.63%). Furthermore, the proposed approach outperforms the baseline reported on the UTA-RLDD secondary dataset (65.20% accuracy). These findings indicate that the proposed framework successfully constructs a representative primary dataset, identifies the optimal feature combination, and comprehensively evaluates the performance of three state-of-the-art deep learning architectures for driver drowsiness classification.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi kantuk, Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), Time Series Transformer, CNN-BiLSTM-Attention, Temporal Convolutional Network (TCN), dataset primer, deep learning, Drowsiness detection, Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), Time Series Transformer, CNN-BiLSTM-Attention, Temporal Convolutional Network (TCN), primary dataset, deep learning
Subjects:	T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.62 Simulation T Technology > T Technology (General) > T57.74 Linear programming
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Tabitha Natasya Cyntia Dewi
Date Deposited:	31 Jul 2026 02:31
Last Modified:	31 Jul 2026 02:31
URI:	http://repository.its.ac.id/id/eprint/140457

Actions (login required)

View Item