Detection of Intoxication in Automobile Drivers

Abstract

The advancement of innovative technology that can accurately aid in the timely detection of intoxication in humans, particularly in scenarios requiring intoxication detection while driving, requires more research attention. Despite the fact that there are conceptual ideas for using facial recognition to perform this task, the results still need improvement. While the existing research works have demonstrated the capabilities of specific architectures, this article intends to develop an improved system to predict whether a driver is intoxicated or not by utilizing an ocular approach and CNNs. It also provides a general overview of face recognition and its applications. It can be assumed that if an improved system is developed and implemented in vehicles, the severe effects of drunk driving will be reduced. This study has tested five methods, within which four methods are composed of CNN architectures: VGG19, VGG16, MobileNet V2, and ResNet 50. The performance of an LSTM+ Attention Mechanism approach is also tested in this scenario. Finally, this article demonstrates that the VGG16 architecture provides the best validation accuracy for the given classification problem while also considering the results of other approaches to assess its applicability in alcoholic detection systems.