Real-Time Driver Distraction Detection System Using Convolutional Neural Networks

Road crashes have emerged as one of the top causes of death among the most productive age group. According to the World Health Organization (WHO), in the last decade 1.3 million people died and 5.3 million people got physically disabled in India due to road accidents. Many of these accidents are preventable because they happen due to driver distractions. In the past, many of the researchers have attempted to solve this problem by using traditional computer vision methods and deep learning to detect driver distractions, but an end-to-end system has not been presented. In this paper, we present a real-time system which performs driver distraction detection using convolutional neural network (CNN) and alerts the driver. The system is implemented using SDC (Software Defined Cockpit) powered by Android IVE (In-Vehicle-Experience). The system is flexible to run on Android smart phones for constrained environments. The detection is performed using the state of the art CNN models like MobileNetV1, MobileNetV2, InceptionV3 and VGG-16. We present the performance analysis of above mentioned CNN models using State Farm Distracted Driver Detection dataset. We extended system implementation and performance analysis of the best performing CNN model amongst the four tested, using our custom dataset to improve the performance for real-time. This custom data set is very close to the end driver scenarios as it covers not only the side view but also the frontal view of the driver, which improves the performance of the system significantly.