陈耀祖, 谷玉海, 成霞, 徐小力. 基于优化YOLOv4算法的行驶车辆要素检测方法[J]. 应用光学, 2022, 43(2): 248-256. DOI: 10.5768/JAO202243.0202003
引用本文: 陈耀祖, 谷玉海, 成霞, 徐小力. 基于优化YOLOv4算法的行驶车辆要素检测方法[J]. 应用光学, 2022, 43(2): 248-256. DOI: 10.5768/JAO202243.0202003
CHEN Yaozu, GU Yuhai, CHENG Xia, XU Xiaoli. Driving vehicle elements detection method based on optimized YOLOv4 algorithm[J]. Journal of Applied Optics, 2022, 43(2): 248-256. DOI: 10.5768/JAO202243.0202003
Citation: CHEN Yaozu, GU Yuhai, CHENG Xia, XU Xiaoli. Driving vehicle elements detection method based on optimized YOLOv4 algorithm[J]. Journal of Applied Optics, 2022, 43(2): 248-256. DOI: 10.5768/JAO202243.0202003

基于优化YOLOv4算法的行驶车辆要素检测方法

Driving vehicle elements detection method based on optimized YOLOv4 algorithm

  • 摘要: 随着车辆数量的急剧增加,带来了一系列管理问题,智能交通系统是一种有效的解决方式。由于传统的目标识别方式受天气、距离、角度、光照等因素的影响较大,且基于原YOLOv4算法的驾驶员面部、手部等信息检测的准确率不高,提出一种基于优化YOLOv4算法的检测定位方法。在给原YOLOv4网络增加一个更小的检测尺度的同时,使用模糊ISODATA动态聚类算法对先验框数目进行优化,并使用真实十字路口数据集进行实验。实验证明,优化后的网络在训练集中的类间平均准确率为98.56%,检测帧频为41.43 帧/s,均高于原网络。

     

    Abstract: The rapid increase in the number of vehicles also brings a series of management problems. The intelligent transportation system is an effective solution. Due to the traditional target recognition method was greatly affected by factors such as weather, distance, angle, and illumination, and the accuracy of the information detection of driver's face, hand which was based on the original YOLOv4 algorithm was not high, a detection and positioning method based on the optimized YOLOv4 algorithm was proposed. While adding a smaller detection scale to the original YOLOv4 network, a fuzzy ISODATA dynamic clustering algorithm was used to optimize the number of a priori frames, and the experiments using the real intersection data set were carried out. The experimental results show that the optimized network has an average accuracy of 98.56% between classes in the training set and a detection frame rate of 41.43, which are higher than those of the original network.

     

/

返回文章
返回