主办:陕西省汽车工程学会
ISSN 1671-7988  CN 61-1394/TH
创刊:1976年

汽车实用技术 ›› 2022, Vol. 47 ›› Issue (18): 28-32.DOI: 10.16638/j.cnki.1671-7988.2022.018.006

• 智能网联汽车 • 上一篇    

基于 Yolo 算法的交通锥标颜色检测

赵梓杉,秦玉英,李 刚,衣明悦   

  1. 辽宁工业大学 汽车与交通工程学院
  • 出版日期:2022-09-30 发布日期:2022-09-30
  • 通讯作者: 赵梓杉
  • 作者简介:赵梓杉(1997—),女,硕士研究生,研究方向为智能驾驶与主动安全,E-mail:1299113812@qq.com。
  • 基金资助:
    辽 宁省科技厅 重大研 发计划 (207106020);辽宁省教育厅项目(JJL201915411);国家自然科学基金 (51605213);辽宁省高等学校国(境)外培养项目(2018LNGXGJWPY-YB014)。

Traffic Cone Color Detection Based on Yolo Algorithm

ZHAO Zishan, QIN Yuying, LI Gang, YI Mingyue   

  1. College of Automobile and Traffic Engineering, Liaoning University of Technology
  • Online:2022-09-30 Published:2022-09-30
  • Contact: ZHAO Zishan

摘要: 为了解决中国大学生无人驾驶方程式大赛的赛车检测交通锥标速度较慢和鲁棒性差的 问题,文章采用自制数据集,提出一种使用自制数据集的 Yolo 实时目标检测方法。针对交通 锥标较为细长、尺寸小的特点,Yolo 使用 K-means 聚类算法对数据集中的真值进行聚类,选 取合适的边界框数量,使目标检测算法融合本数据集的类别并实现锥桶检测以及三种颜色的 分类。实验结果表明,在不同的外界环境中,Yolov5 网络的交通锥标颜色分类检测模型的检 测准确率高、鲁棒性好、计算速度快。在少量数据的情况下召回率达到 88.84%,准确率达到 86.87%,比 Yolov3 算法提高了 36.78%,比原始算法提高了 44.8%,检测速度(34 f/s)满足 赛事需求。

关键词: 颜色检测;Yolov5 网络;Yolo 算法;交通锥标

Abstract: In order to solve the problem of slow speed and poor robustness of formula student China traffic cone mark detection, this paper proposes a real-time target detection method of Yolo using self-made data set. Aiming at the characteristics of traffic cone sign is relatively thin and small in size, K-means clustering algorithm is used to cluster the true value of data set, and the appropriate number of boundary boxes is selected to make the target detection algorithm integrate the classification of this data set and realize cone detection and classification of three colors. The experimental results show that the traffic cone color classification detection model of Yolov5 network in different external environment, the detection accuracy is high. The algorithm is robust and fast to calculate. In the case of a small amount of data, the recall rate reaches 88.84%, the accuracy rate reaches 86.87%, 36.78% higher than Yolov3 algorithm, 44.8% higher than the original algorithm, and the detection speed (34 f/s) meets the requirements of the event.

Key words: Color detection; Yolov5 network; Yolo algorithm; Traffic cone