Implementation of Scale-Invariant Feature Transform Convolutional Neural Network for Detecting Distracted Driver

Abstract

A distraction while driving a vehicle may result in fatal consequences, namely accidents that may leave road users seriously injured or even dead. In order to mitigate this risk, it is imperative to establish a distracted driver detection system that is both precise and real-time. This research focuses on the application of artificial intelligence, with a particular emphasis on deep learning, which is achieved through the utilization of the Convolutional Neural Network (CNN) model. In order to enhance the detection of inattentive drivers and produce a more precise model, a scaleinvariant feature transform (SIFT)-CNN combination is proposed. The activities of the driver while operating a vehicle are categorized into ten categories in this study. One of these categories is considered a normal condition, while the remaining nine are classified as inattentive behaviors. This study implemented Adam optimization with 64 batches, a learning rate of 0.001, and epochs of 20, 25, 50, and 100. The proposed CNNSIFT model is capable of achieving superior performance in comparison to the solitary CNN model, as evidenced by the experimental results. The CNN-SIFT model has achieved 99% accuracy and a 0.05 loss when the hyperparameter configuration is optimized for 50 epochs. The analysis indicates that the accuracy of the features obtained from CNN-SIFT can be improved by approximately 1% compared with CNN to classify the type of driver distraction behavior. The model's reliability was further enhanced by its evaluation on test data, which resulted in high accuracy, precision, recall, and F1-score values. The model's ability to accurately identify driver behavior with a high degree of reliability is demonstrated by these results, which are a positive contribution to the improvement of road safety.