DEVELOPMENT OF DRIVER DROWSINESS DETECTION ALGORITHM
This project proposes two different non-intrusive approaches to detect driver drowsiness to ensure the safety of the drivers and road users. Psychophysiological-based measurement is not feasible in practice as it causes driving distraction and inconvenience for the drivers by wearing special equipme...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project Report |
| Language: | English English |
| Published: |
Universiti Malaysia Sarawak, (UNIMAS)
2022
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/40129/1/Yvonne%20Phua%20Yee%20Wun%2024pgs.pdf http://ir.unimas.my/id/eprint/40129/5/Yvonne%20Phua%20Yee%20Wun%20ft%20%281%29.pdf http://ir.unimas.my/id/eprint/40129/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This project proposes two different non-intrusive approaches to detect driver drowsiness to ensure the safety of the drivers and road users. Psychophysiological-based measurement is not feasible in practice as it causes driving distraction and inconvenience for the drivers by wearing special equipment on the body. Some studies that use computer vision techniques only focus on the eyes to detect drowsiness, which leads to limitations for drivers with smaller eyes and with sunglasses. To contribute to the existing works of driver drowsiness detection systems, this project aims to develop a more useful drowsiness detection algorithm with low complexity and high performance using Python 3.10.1 software. The first proposed method uses facial landmarks to identify blinks and yawns based on suitable thresholds set for the drivers. The second approach applies deep learning methods with three different convolution neural network models, which are modified LeNet-5, MobileNet-V2, and DenseNet-201 to detect drowsiness. Two public video datasets are used to test the proposed algorithms, namely Yawning Detection Dataset (YawDD) and Driver Drowsiness Dataset (D3S). The deep learning approaches perform better than the technique that calculates the eye and mouth aspect ratios to detect drowsiness. The modified LeNet-5 achieved the highest accuracy of 92.22% among the four proposed algorithms. It sets a great benchmark for future work on driver drowsiness detection. This research has provided meaningful solutions to prevent drowsy driving accidents. |
|---|