Scholarlink

The rapid advancement of technology has significantly impacted autonomous vehicle systems, which are crucial for effective smart traffic management. Potholes, caused by adverse weather, heavy vehicle loads, and traffic accidents, are a leading contributor to vehicular mishaps. This risk is particularly pronounced during nighttime when human visibility is severely limited, highlighting the necessity of real-time object detection algorithms to identify potholes both at close range and from a distance, thereby enhancing traffic safety. This study demonstrates that the YOLO family of single-stage object detection models is well suited for real-time applications due to its impressive performance and speed efficiency. A comprehensive dataset of pothole images was developed using cameras mounted on vehicles and unmanned systems. The YOLOv7 and YOLOv8 models were subjected to training, validation, and testing using this dataset. Results indicated that the YOLOv8 model outperformed YOLOv7 in terms of durability for real-time applications. This study thoroughly analyzes the strengths and weaknesses of both models under challenging conditions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of YOLOv7 and YOLOv8 Models for Real-Time Pothole Detection in Autonomous Driving

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hilal Kuzu,
  • Murat Bakirci

摘要

The rapid advancement of technology has significantly impacted autonomous vehicle systems, which are crucial for effective smart traffic management. Potholes, caused by adverse weather, heavy vehicle loads, and traffic accidents, are a leading contributor to vehicular mishaps. This risk is particularly pronounced during nighttime when human visibility is severely limited, highlighting the necessity of real-time object detection algorithms to identify potholes both at close range and from a distance, thereby enhancing traffic safety. This study demonstrates that the YOLO family of single-stage object detection models is well suited for real-time applications due to its impressive performance and speed efficiency. A comprehensive dataset of pothole images was developed using cameras mounted on vehicles and unmanned systems. The YOLOv7 and YOLOv8 models were subjected to training, validation, and testing using this dataset. Results indicated that the YOLOv8 model outperformed YOLOv7 in terms of durability for real-time applications. This study thoroughly analyzes the strengths and weaknesses of both models under challenging conditions.