北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2022, Vol. 45 ›› Issue (5): 103-108.

• 研究报告 • 上一篇    下一篇

改进YOLOv3算法的交通多目标检测方法

宋宇博,高嘉振   

  1. 兰州交通大学 机电技术研究所
  • 收稿日期:2021-10-12 修回日期:2022-01-23 出版日期:2022-10-28 发布日期:2022-11-01
  • 通讯作者: 高嘉振 E-mail:ggaojiazhen@163.com
  • 基金资助:
    青年博士基金项目

Improved YOLOv3 Algorithm for Multi-Target Detection of Traffic

SONG Yubo, GAO Jiazhen   

  • Received:2021-10-12 Revised:2022-01-23 Online:2022-10-28 Published:2022-11-01

摘要: 针对交通场景中目标像素占比小互相遮掩等因素造成漏检误检的问题,提出了基于 YOLOv3 的多目标检测方法该方法在 YOLOv3 网络结构中植入空间金字塔池化模块以增强特征表达,同时提出一种多尺度特征融合机制兼顾获取空间信息和语义信息,通过扩展预测层的预测分支来细化待检目标的语义信息此外,将改进的 K均值聚类算法用于提取先验框的初始中心点,提升预测锚框与待检目标的匹配度,并运用柔性非极大值抑制算法进行置信分数的灵活调整基于混合数据集的实验结果表明,所提方法有效地提升了检测精度

关键词: 交通多目标检测, 特征提取, 多尺度特征融合, 预测锚框匹配

Abstract:

A YOLOv3-based multi-target detection method is proposed to address the problem of missed and false detection caused by the small percentage of target pixels and mutual occlusion in traffic scenes. The method implants a spatial pyramid pooling module in the YOLOv3 network structure to enhance feature representation, and a multi-scale feature fusion mechanism is proposed to obtain both spatial information and semantic information. The semantic information of the target to be detected is refined by extending the prediction branch of the prediction layer. In addition, the improved K-means + + clustering algorithm is used to extract the initial center of the priori box and improve the matching degree between the prediction anchor box and the target to be detected. Meanwhile, a flexible non-maximum suppression algorithm is applied to adjust the confidence score flexibly. The experimental results based on the hybrid data set show that the proposed method improves the detection accuracy effectively.

Key words: traffic multi-target detection,  feature extraction,  multiscale feature fusion,  predictive anchor box matching

中图分类号: