A Self-Supervised Approach for Detection and Analysis of Driving Under Influence

Le résumé fourni par la source

Driving under the influence (DUI) of alcohol or drugs has consistently been among the primary causes of traffic accidents. Current DUI detection models typically rely on supervised learning, necessitating extensive annotated DUI data. In this paper, we propose a self-supervised approach that combines the variational autoencoder and the isolation forest. Our method not only detects DUI events but also can provide explanations for them. We collected driving data from multiple drivers in both normal and drunk states in simulated urban Japanese environments. Experimental results on this dataset indicate that our method achieves comparable performance to other supervised learning methods, even when utilizing only normal data.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.