Poster Abstract: A Low-cost Driving Risk Prediction System for Hilly Roads via Surveillance Cameras and OnsiteWebcams

Abstract

Smart transportation is a potential research area that aims to make the roads safer to drive, mainly in hilly areas where roads are curvy, narrow, and dangerous. It uses predictive power of artificial intelligence to identify road risks. These risks can be predicted by the use of surveillance cameras and onsite webcams to have a safer drive. However, generating road risk descriptions using traditional models need high computational costs. Therefore, we propose a low-cost, lightweight driving risk prediction system for hilly roads that uses the meta-learning concept to generate accurate and distinctive descriptions. Meta-learning utilizes both supervised and reinforcement concepts by learning simultaneously to obtain a global optimal solution on our RoadCaption dataset. The experiments show that our system generates captions that predict the road conditions of hilly areas with high accuracy. © 2022 ACM.