Download PDFOpen PDF in browserInattentional Driver Detection Using Faster R-CNN and ResNetEasyChair Preprint 1320514 pages•Date: May 7, 2024AbstractAddressing the critical issue of inattentional driving, this paper proposes an innovative method combining the Faster R-CNN (Region-based Convolutional Neural Network) and ResNet (Residual Neural Network) architectures. Detecting inattentional driving behaviors remains challenging due to the variability in driver behavior and the subtle nature of distractions. By integrating Faster R-CNN's object detection capabilities with ResNet's robust feature extraction, our approach aims to improve detection accuracy. Utilizing diverse datasets gathered from real-world driving scenarios, our method ensures the model's adaptability to different driving conditions and distractions. Furthermore, we employ a saliency technique to identify regions of interest within the driver's field of view, guiding the model to focus on crucial areas indicative of inattentional behaviors. It also uses the perinasal perspiration and cognitive senses for more evaluation purpose.It uses the IOU as evaluation metrics. A key advantage of our approach lies in its efficiency in handling large datasets without sacrificing performance. The ResNet architecture's deep residual learning framework enables effective feature extraction, while Faster R-CNN facilitates precise object detection within these features. This integration ensures the accurate identification of inattentional cues while maintaining computational efficiency. Through rigorous evaluation of diverse datasets, our approach demonstrates promising results in accurately detecting inattentional driving behaviors. By reducing misdiagnoses and enabling timely interventions, our method enhances road safety and prevents accidents caused by driver distraction. This research represents a significant advancement in leveraging deep learning techniques for effective inattentional driving detection, with potential implications for global road safety. Keyphrases: Faster R-cnn(Faster Region-Convolutional Neural Network), IOU(Intersection over union), Resnet(Residual Network)
|