多目标小尺度车辆目标检测方法的研究
作者:
作者单位:

1.哈尔滨理工大学电气与电子工程学院;2.哈尔滨工程大学

作者简介:

通讯作者:

中图分类号:

TP391.4

基金项目:

国家自然科学基金项目(面上项目,重点项目,重大项目)


Research on Multi-target and Small-scale Vehicle Target Detection Method
Author:
Affiliation:

1.Harbin University of Science and Technology;2.Harbin Engineering University

Fund Project:

The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    车辆目标检测是智能交通系统中的重要环节,针对传统车辆目标检测方法效率低,小目标检测效果不好,漏检率高等问题,提出了一种基于改进的YOLOv3网络车辆目标检测算法。为了提高了车辆检测的效率,利用轻量化模型MobileNet v2代替原YOLOv3中的特征提取网络,使得网络计算量相比原算法有所降低。为了有效提高网络对小尺度车辆目标的检测能力,网络将由高到低不同尺度的特征层融合之后进行目标检测。为了得到更丰富的语义特征信息,提高网络的预测能力,增加了特征增强模块。同时针对车辆目标检测的特定应用,利用K-means方法对锚框重新聚类以满足车辆目标检测的特定需求。结合以上的改进获得了车辆目标检测网络YOLOv3-M2,实验结果表明,与YOLOv3相比,改进的方法平均检测准确率增加了约9%,时间减少约一半,不仅提高了检测效率,同时提高了小目标检测能力。

    Abstract:

    Vehicle target detection is an important link in intelligent transportation system. Aiming at the problems of low efficiency, poor detection effect of small targets and high miss rate of traditional vehicle target detection methods, a vehicle target detection algorithm based on improved YOLOv3 network is proposed. In order to improve the efficiency of vehicle detection, the lightweight model MobileNet v2 is used to replace the feature extraction network in the original YOLOv3, and the calculation amount is reduced compared with the original algorithm. In order to effectively improve the network"s ability to detect small-scale vehicle targets, feature layers of different scales are fused and target detection is carried out on feature maps of different scales. At the same time, in order to obtain more abundant semantic feature information and improve the prediction ability of network, a feature enhancement module is proposed. For the specific application of vehicle target detection, K-means is used to re-cluster anchor frames to meet the requirements of vehicle target detection. Combined with the above improvements, the vehicle target detection network YOLOv3-M2 is obtained. Experimental results show that compared with YOLOv3, the improved method not only improves the detection efficiency, but also improves the small target detection capability, increasing the average detection accuracy of the network by about 9%.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-05-26
  • 最后修改日期:2020-08-03
  • 录用日期:2020-08-04
  • 在线发布日期:
  • 出版日期: