YOLOv5 및 MiDaS 기반 포트홀 실시간 탐지 및 깊이 추정 시스템 = Real-Time Pothole Detection and Depth Estimation System Based on YOLOv5 and MiDaS

다국어 초록 (Multilingual Abstract)

Due to recent climate change, increased traffic volume, and the aging of road infrastructure, the occurrence of potholes has been on the rise, posing significant risks to driver safety and vehicle operation. Accordingly, the importance of technologies that can detect and repair potholes quickly and accurately is gaining attention, as they are recognized as a key task for ensuring traffic safety and improving the efficiency of road maintenance. Conventional pothole detection technologies have evolved based on vibration sensors, laser measurement, and image recognition; however, most of them are limited to detecting the mere presence of potholes without providing quantitative information on their depth or risk level. To address this issue, this paper proposes a deep learning-based real-time pothole detection system that combines YOLOv5-based pothole detection with a depth estimation model. Furthermore, by utilizing a Large Language Model(LLM) to generate user-friendly warning messages and delivering them via voice guidance, the system enables drivers to immediately recognize dangers without relying on visual information.