基于改进 YOLO11n的大车盲区人员检测方法
CSTR:
作者:
作者单位:

(1.重庆邮电大学通信与信息工程学院,重庆 400065;2.昆明云内动力股份有限公司,昆明 650200)

作者简介:

代少升(1974—),男,河南省潢川县人,博士,教授,主要从事红外成像系统及 SOPC嵌入式系统设计与开发;余自安(1974—),男,河南省光山县人,高级工程师,主要从事机械设计及理论研究。

通讯作者:

中图分类号:

TP391

基金项目:


Personnel Detection in Large Vehicle Blind Spots Using Improved YOLO11n
Author:
Affiliation:

(1. School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications, Chongqing 400065,CHN;2. Kunming Yunnei Power Co.,Ltd.,Kunming 650200,CHN)

Fund Project:

  • 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:

    针对大车盲区中行人和骑行者目标检测存在遮挡、检测精度低、实时性差等问题,文章提出一种基于改进 YOLO11n网络的盲区人员检测方法。首先,在主干网络中引入感受野增强注意力卷积模块(RFCBAMConv),改进 C3k2模块,以提升对遮挡目标的感知能力;其次,引入内容引导注意力特征融合结构(CGAFusion),引导网络关注关键区域特征,增强多尺度语义融合效果;最后,采用多尺度特征共享检测头(MFSD),通过多分支共享卷积核降低模型参数量与计算负担。实验在自建大车盲区数据集上进行,结果表明所提方法在 mAP@0. 5上分别提升约 2. 5%,召回率显著高于 YOLOv5n等同类轻量模型,同时保持仅 2.87 M参数量和 8. 2 GFLOPs的低计算复杂度,验证了其在精度、效率与部署适应性上的优势。该方法适用于资源受限的边缘计算设备,在智能交通安全领域具有广阔的应用前景。

    Abstract:

    This paper presents a blind spot detection method based on an improved YOLO11n network that addresses the challenges of pedestrian and cyclist detection in large vehicle blind spots, including occlusion,low detection accuracy,and poor real-time performance. First,the receptive-field augmented attention convolutional module(RFCBAMConv)is introduced into the backbone network to improve the C3k2 module and thus enhance the perception of occluded objects. Second,the content-guided attention feature fusion architecture(CGAFusion)is incorporated to ensure that the network focuses on key regional features,enhancing multiscale semantic fusion. Finally,a multiscale feature sharing detection (MFSD)head is employed,which ensures that convolution kernels are shared across multiple branches and thus reduces the number of model parameters and computational burden. Experiments on a self-constructed large vehicle blind spot dataset demonstrate that the proposed method achieves approximately 2. 5% improvement in mAP@0. 5 and significantly higher recall compared with similar lightweight models such as YOLOv5n,while maintaining low computational complexity(only 2. 87 M parameters and 8. 2 GFLOPs). These results demonstrate its advantages in accuracy, efficiency, and deployment adaptability. Therefore,the proposed method is suitable for resource-constrained edge computing devices and has broad application prospects in the field of intelligent transportation safety.

    参考文献
    相似文献
    引证文献
引用本文

代少升,宋国庆,余自安.基于改进 YOLO11n的大车盲区人员检测方法[J].半导体光电,2025,46(6):1071-1078. DAI Shaosheng, SONG Guoqing, YU Zian. Personnel Detection in Large Vehicle Blind Spots Using Improved YOLO11n[J].,2025,46(6):1071-1078.

复制
分享
相关视频

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:2025-10-09
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2025-12-26
  • 出版日期:
文章二维码