IntoxDetectV2: Comparative Analysis of CNNs and LLMs for Detection of Intoxication through Ocular and Facial Features Using Ensemble Learning

Drunk driving remains a significant cause of fatalities in the U.S., with one person dying every 39 minutes. Current intoxication detection methods are expensive, time-consuming, and often require specialized equipment and trained personnel [1]. We propose a non-invasive approach using facial image analysis to detect intoxication. Our results show that the VGG16 model, optimized with Adamax and trained on 10,000 facial images, achieved 83% accuracy in detecting inebriation. This method demonstrates strong potential as an AI-based facial intoxication detection system, offering a practical solution to combat drunk driving.