Distracted Driver Detection on Re-trained ResNet Architecture
Vinay Thandra, Arun Soppimath, Ravi Kant Kumar, Writankar Kundu, Piyush Telele, Vaishali Balaji, Narayana Darapaneni, Anwesh Reddy Paduri · 2023
Road accidents are predominantly caused by the distracted drivers and nearly 1.3 million deaths are automobile accidents, of which, drivers are held responsible for 78% of accidents.There are various reasons for driver distractions which are drinking, operating instruments, mobile usage, interacting with fellow passengers etc.For the scope of this project, we intend to develop a model to successfully identify whether thedriver is driving safely or is distracted using a combined dataset from the State Farm Distracted Driver Detection challenge on Kaggle& the AUC (American University in Cairo) Project.Convolutional Neural Network with ResNet architecture was used in developing the model.Grad-CAM technique was used to identify gradients in parts of images which impacted classification of images.Explainable AI can help build better models.This approach was able to provide us with promising accuracy and definite results.Experimental results show that our system achieves an accuracy of 99% on the Kaggle dataset and82% on the AUC data set.