الفهرس | Only 14 pages are availabe for public view |
Abstract This study discusses the problem of driver distractions and how to mitigate it by detecting driver distractions using deeplearning models. The thesis describes all the steps done starting from collecting the dataset till building, training, and analyzing the models. The thesis organized as follows: Chapter 1 starts with talking about road accidents and how driver distractions are a main part of the problem. Then the chapter discusses the shortage of the existing driver distractions datasets and models due to the lack of diversity of lighting conditions. Finally, this chapter presents the contributions of this research to treat this shortage by recording a driver distractions dataset that represents the real driving experience and train seven models using the collected data. Chapter 2 discusses the theoretical background of this thesis and presents the required important concept for this research understanding. This chapter starts with presenting the concept of NoIR camera. Then the chapter discusses the deeplearning concepts. Chapter 3 provides a comprehensive study about the existing studies in the literature of driver distraction recognition. Then discusses the contribution and shortage of each study. Chapter 4 is concerned with the methodology and the steps of the two main parts of this research, dataset collection and models training. In the first part, this chapter introduces the steps of dataset collection and annotation. In the second part, the methodology of building and training the deeplearning models for driving distractions recognition is discussed. Chapter 5 shows the experimental results of the dataset and the trained models. For the dataset, this chapter presented the dataset analysis and compares this dataset with other datasets. For the trained models, this chapter shows the accuracy of each model and the analysis of these models using different techniques like confusion matrix, t-SNE representation, and saliency map. Chapter 6 presents the conclusion of the thesis in the first part of it. In the second part, this chapter introduces the suggestions for future extensions related to the enhancement of the dataset, the trained models, and applications. |