基于双主干网络的分心驾驶行为检测方法
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嵌入式系统的设计与开发;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. 41

基金项目:


Dual-Backbone Network-Based Method for Distracted Driving Behavior Detection
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:

    针对现有分心驾驶行为检测算法检测精度低、计算量大等问题,文章提出一种基于 YOLOv11n改进的双主干网络(YOLO-DCC)。首先,该算法使用两条不同的轻量化主干替换 YOLOv11n的原始主干,在保持轻量化的同时,增强了模型的特征提取能力。随后,引入跨注意力融合模块(CAFM)在通道与空间维度上实现跨分支特征的动态交互与加权融合,进一步提升了关键特征的表达能力。最后,在颈部网络中引入上下文引导融合模块(CGFM),通过聚合局部、周边与全局上下文信息,优化了多尺度特征的融合效果,提升了模型的检测精度。在 StateFarm数据集上进行实验,结果表明所提算法将 mAP@0. 5提升到了 95. 7%,相比于 YOLOv11n算法检测精度提高了 3. 1%,计算量减少了 7. 94%,能够有效检测驾驶员的分心驾驶行为。

    Abstract:

    To overcome the limitations of low detection accuracy and high computational load in existing distracted driving detection algorithms,in this study,an improved model based on the YOLOv11n architecture—namely, YOLO-DCC—is proposed. First, the algorithm replaces the original YOLOv11n backbone with two distinct lightweight backbones,which enhance its feature extraction capabilities while maintaining a lightweight architecture. Subsequently,the cross-attention fusion modules(CAFMs)are employed to achieve dynamic interaction and weighted fusion of cross-branch features along both channel and spatial dimensions,and thus enhance the representation of key features. Finally,context guided fusion modules(CGFMs)are introduced into the neck network to optimize the fusion of multiscale features via the integration of local, surrounding, and global contextual information,which improve the detection accuracy of the model. Experiments performed on the StateFarm dataset show that the proposed algorithm achieves a mean average precision(mAP) of 95. 7% at an IoU threshold of 0. 5 while reducing the computational load by 7. 94%. This mAP represents a 3. 1% improvement over that of the baseline YOLOv11n model,demonstrating the effectiveness of the proposed model in detecting various distracted driving behaviors.

    参考文献
    相似文献
    引证文献
引用本文

代少升,何帅兵,余自安.基于双主干网络的分心驾驶行为检测方法[J].半导体光电,2026,47(1):172-179. DAI Shaosheng, HE Shuaibing, YU Zian. Dual-Backbone Network-Based Method for Distracted Driving Behavior Detection[J].,2026,47(1):172-179.

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:2025-11-14
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-03-20
  • 出版日期:
文章二维码