Scholarlink

This work focuses on the application of a model implementing YOLO (You Only Look Once) that can be applied to ADAS (Advanced Driver Assistence Systems) in order to improve the safety and autonomy of these systems. The validation of this model is oriented to the identification of traffic signs in images carried out in a 1:10 scale autonomous driving vehicle through the use of a camera, allowing the system to capture the vehicle environment in real time and process the images to detect relevant signs such as stop, left turn, right turn, dead center, and no entry. With this, a total of 3000 images were captured in different positions and angles of each of the signs, obtaining a Database of a total of 15000 images for a robust training. The training was carried out in different versions of YOLO in this case comparing the latest versions YOLO v11n which was designed to improve the speed of inference without compromising accuracy, and YOLO v8s a lightweight and compact version with high computational efficiency; in both cases we could see which has better performance in the training applied to sign detection. However, YOLOv11n showed a slight advantage in terms of speed of inference. Both models achieved identical metrics (mAP50-95 = 0.99, accuracy and recall = 1.00), however YOlOv11n shows a superior runtime efficiency with a total 120 ms latency per image compared to 447 ms for YOLOv8s.

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

Traffic Sign Detection for Autonomous Driving at 1:10 Scale a YOLO Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Luis Antonio Márquez Carlos,
  • Erika P. Sanchez Femat,
  • Huitzilopoztli Luna Garcia,
  • Rafael Reveles Martinez,
  • Jose Guadalupe Arceo Olague,
  • Juvenal Villanueva Maldonado,
  • Juan Ruben Delgado Contreras,
  • Manuel Alejandro Soto Murillo,
  • Jorge Isaac Galvan Tejada,
  • Jose Maria Celaya Padilla

摘要

This work focuses on the application of a model implementing YOLO (You Only Look Once) that can be applied to ADAS (Advanced Driver Assistence Systems) in order to improve the safety and autonomy of these systems. The validation of this model is oriented to the identification of traffic signs in images carried out in a 1:10 scale autonomous driving vehicle through the use of a camera, allowing the system to capture the vehicle environment in real time and process the images to detect relevant signs such as stop, left turn, right turn, dead center, and no entry. With this, a total of 3000 images were captured in different positions and angles of each of the signs, obtaining a Database of a total of 15000 images for a robust training. The training was carried out in different versions of YOLO in this case comparing the latest versions YOLO v11n which was designed to improve the speed of inference without compromising accuracy, and YOLO v8s a lightweight and compact version with high computational efficiency; in both cases we could see which has better performance in the training applied to sign detection. However, YOLOv11n showed a slight advantage in terms of speed of inference. Both models achieved identical metrics (mAP50-95 = 0.99, accuracy and recall = 1.00), however YOlOv11n shows a superior runtime efficiency with a total 120 ms latency per image compared to 447 ms for YOLOv8s.