A Framework for Instantaneous Driver Drowsiness Detection Based on Improved HOG Features and Naïve Bayesian
Samy Bakheet1,2, Ayoub Al-Hamadi2
1Department of Information Technology, Faculty of Computers and Information, Sohag University, P. O. Box 82533 Sohag, Egypt.
Brain Sciences
|March 6, 2021
Summary
This study introduces an improved Histogram of Oriented Gradient (HOG) method for driver drowsiness detection. The novel framework achieves 85.62% accuracy, offering an efficient and stable solution.
Area of Science:
- Computer Vision
- Machine Learning
- Human-Computer Interaction
Background:
- Histogram of Oriented Gradient (HOG) features are widely used in computer vision due to their distinctiveness and robustness.
- Driver drowsiness is a significant safety concern, necessitating reliable detection methods.
Purpose of the Study:
- To propose an innovative framework for driver drowsiness detection using an improved HOG feature descriptor.
- To enhance the distinctiveness, robustness, and compactness of HOG features for this specific application.
Main Methods:
- An adaptive descriptor was developed from an improved HOG feature based on binarized histograms of shifted orientations.
- The generated descriptor was fed into a trained Naïve Bayes (NB) classifier for drowsiness determination.
Main Results:
- The proposed framework achieved a competitive detection accuracy of 85.62% on the NTHU-DDD dataset.
- The method demonstrated efficiency and stability, comparable to state-of-the-art baselines.
Conclusions:
- The developed framework shows potential as a strong contender for driver drowsiness detection systems.
- The improved HOG features offer a promising approach for enhancing the accuracy and reliability of drowsiness detection.
