Inattentive Driver Identification Smart System (IDISS)

Every year, distracted driving is one of the leading causes of fatalities on the road. Distracted driving practices may include texting, eating, drinking, or other such interruptive actions that may shift the focus of the attention of the driver while driving. Several measures have been undertaken to identify and prevent such mishaps, and many authors have put forth deep learning techniques to recognize these driving patterns. Taking note of it, this study analyzes the proposed systems and suggestions made by other authors to mitigate distracted driving and aims to propose a hybrid system that would be trained to recognize the distracted action of the driver in the car with the aid of deep learning models and alert the driver about the same with the help of IoT devices. The proposed sytem is capable of handling multiple tasks simultaneously with an optimized Deep Learning model and IoT devices. The ultimate goal of this paper is to propose a system that deploys a transfer learning approach using pre-trained deep learning models to produce a highly accurate classification result. The accuracy of determining whether the driver is distracted or not will be impacted using the transfer learning concept and fine-tuning on the pre-trained models like VGG16, VGG19, Xception, ResNet-50, and InceptionV3. A sensor and microprocessor would simultaneously work to detect the actions of the driver, classify the data, and would then deliver an appropriate reaction to the alarm system. The alert system consists of a display system which will accept data from the classification algorithm and will inform the driver of any distracted driving tendencies. When acts of distracted driving are identified, the alarm system will notify the driver. Among the deep learning models trained and tested, ResNet-50 gives the highest validation accuracy of 99.8% in classifying whether the driver is distracted or not. The microprocessor takes this result and sends it to the alert system.