PROJECT TITLE :

Distracted Driver Detection Based on a CNN With Decreasing Filter Size

ABSTRACT:

The number of people who have lost their lives in car accidents because they were distracted while driving has seen a sharp rise in recent years. Deep Learning technology, which is rapidly developing, currently has the capability to identify instances of distracted driving. In spite of this, real-time detection is essential, which means that an optimized network must satisfy three requirements that seem to be in conflict with one another: a limited number of parameters, a high level of accuracy, and a high rate of processing. We propose a new D-HCNN model that is based on a decreasing filter size and has only 0.76M parameters. This is a significantly smaller number of parameters than the models that were used in the majority of the other studies. The performance can be improved with D-HCNN through the utilization of HOG feature images, L2 weight regularization, dropout, and batch normalization. AUC Distracted Driver (AUCD2) and State Farm Distracted Driver Detection are both public datasets, and we conduct experimental evaluations on both of them. We discuss the benefits and principles of D-HCNN in detail, and we present our findings (SFD3). The accuracy achieved by AUCD2 and SFD3 is higher than the accuracy achieved by a great number of other state-of-the-art methods, coming in at 95.59% and 99.87%, respectively.