Scholarlink

Deep learning has significantly enhanced object detection capabilities in Advanced Driver Assistance Systems (ADAS), enabling more accurate and timely decision making. This study presents a systematic comparison of three recent YOLO detector generations YOLOv7, YOLOv8, and YOLOv9 focused on traffic sign detection. While maintaining computational efficiency suitable for real-time applications, each generation incorporates architectural innovations that improve detection accuracy and speed. Evaluated on a custom traffic sign dataset, our findings reveal notable performance differences: YOLOv9 (GELAN-C) achieves a mean Average Precision (mAP) of 99% with 32 frame per second (FPS), YOLOv7 achieves 90% mAP with 89 FPS, and YOLOv8n reaches 98% mAP with 167 FPS. This evaluation empowers users to select the most appropriate model based on their embedded platform’s computational constraints and application requirements. By understanding the trade-offs between accuracy and inference speed, developers can tailor traffic sign detection solutions to the specific capabilities of their hardware, ensuring optimal performance and reliability in diverse real-world environments ranging from resource limited devices to high performance systems.

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

Deep Learning Models for Traffic Sign Detection: An Empirical Comparison of YOLO Variants

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Imane Taouqi,
  • Abdessamad Klilou,
  • Assia Arsalane,
  • Kebir Chaji

摘要

Deep learning has significantly enhanced object detection capabilities in Advanced Driver Assistance Systems (ADAS), enabling more accurate and timely decision making. This study presents a systematic comparison of three recent YOLO detector generations YOLOv7, YOLOv8, and YOLOv9 focused on traffic sign detection. While maintaining computational efficiency suitable for real-time applications, each generation incorporates architectural innovations that improve detection accuracy and speed. Evaluated on a custom traffic sign dataset, our findings reveal notable performance differences: YOLOv9 (GELAN-C) achieves a mean Average Precision (mAP) of 99% with 32 frame per second (FPS), YOLOv7 achieves 90% mAP with 89 FPS, and YOLOv8n reaches 98% mAP with 167 FPS. This evaluation empowers users to select the most appropriate model based on their embedded platform’s computational constraints and application requirements. By understanding the trade-offs between accuracy and inference speed, developers can tailor traffic sign detection solutions to the specific capabilities of their hardware, ensuring optimal performance and reliability in diverse real-world environments ranging from resource limited devices to high performance systems.