TRIPLE STAGE CNN MODEL TO DETECT THE DROWSINESS AMONG THE DRIVER’S BASED ON FACE POSITIONING TECHNIQUE

Abstract

Generally, we need strong detection tools to combat driver drowsiness, which is an increasing problem around the world. Deep Learning (DL) and Machine Learning (ML) methods, such as different types of Convolutional Neural Networks (CNNs), have been tried by researchers, but they haven't been able to be used in real-time situations because they are too hard to implement because they are so complicated to compute and don't work very well when tested. Regarding these restrictions, a multistage adaptive multi-level -CNN model has been developed specifically for Driver Drowsiness Detection (DDD), emphasizing the resolution of system complexity and performance challenges. The suggested system employs a multiple tier methodology for driver sleepiness detection, incorporating breakthroughs from multiple separate research stages. The findings illustrate the efficacy of the suggested model across various datasets. The proposed model outperforms previous models, such as YOLOv7-Face, MTCNN, AAMS, and Hyperface-Resnet, in terms of bounding box regression and facial point localization accuracy. The results demonstrate the proposed model's efficacy in facial positioning and the detection of initial driver drowsiness, rendering it helpful for improving road safety.