Facial Intoxication Detector with AI

The purpose of this study is to analyze whether Artificial Intelligence can be used to more effectively detect and prevent drunk driving and if a Machine Learning models like Logistic Regression and Decision Tree can be more accurate than a human police officer.

Drunk driving is a very widespread problem, causing many casualties and millions of dollars in insurance and damages per year. In 2020, despite the COVID19 pandemic greatly reducing road traffic, about 32 people died per day in the US alone from DUI (Driving under Influence)(nhtsa.gov). Existing solutions such as police testing drivers suspected to be under influence is simply impractical considering the number of possible intoxicated drivers. With more and more people gaining access to a vehicle, it is crucial that more effective strategies be developed to detect and combat drunk driving. The purpose of this study is to analyze whether Artificial Intelligence can be used to more effectively detect and prevent drunk driving and if a Machine Learning models like Logistic Regression and Decision Tree can be more accurate than a human police officer. To address this question, a dataset named drunkImagesWebp was used. Two machine learning algorithms, Logistic regression and decision tree were then trained on this dataset with facial image data of intoxicated people to accurately predict the sobriety of humans based on facial cues. After testing this model, it became clear that both Logistic regression and decision tree models can indeed accurately test a driver for signs of intoxication with well over 90% accuracy compared to human-administered tests with can only hit up to around 75% accuracy. By comparison, both the Logistic regression and Decision tree algorithms detected intoxication with 96% accuracy. This paper shows the potential AI can have in creating an automated solution to detecting and ultimately preventing drunk driving.

Explore More!

Research Paper