层内调节特征金字塔的密集人群姿态估计算法
CSTR:
作者:
作者单位:

(1. 华北理工大学 电气工程学院, 河北 唐山 063210;2. 唐山市数字媒体工程技术研究中心, 河北 唐山 063210)

作者简介:

谷学静(1972-),女,河北省唐山市人,博士,教授,主要研究方向为数字媒体技术、人机交互、虚拟人及智能Agent、人工心理和人工情感;

通讯作者:

中图分类号:

TP391.4

基金项目:

河北省自然科学基金高端钢铁冶金联合研究基金专项项目(F2017209120);唐山市沉浸式虚拟环境基础创新团队项目(18130221A).通信作者:郭志斌


Dense Crowd Pose Estimation Algorithm for In-layer Adjustment Feature Pyramid
Author:
Affiliation:

(1. College of Electrical Engineering, NorthChina University of Science and Technology, Tangshan 063210, CHN;2. Tangshan Digital Media Engineering Technology Research Center, Tangshan 063000, 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:

    针对现有姿态估计算法对密集人群检测存在漏检和误检等问题,提出一种改进YOLOv8sPose的密集人群姿态估计算法YOLOv8Pose-DC。首先,设计一种集中式层内调节特征金字塔网络,采用并联方式把可变形注意力机制和CASPPF结合起来,通过自上而下的方式对金字塔网络进行全局集中调节,增加网络中全局表示的空间权重,使得改进算法能够获得全面且具有区分性的特征表示;其次,提出多尺度双检测头结构,减少计算量的同时提高模型检测效率;然后,使用DySample模块,提高模型上采样效率;最后,加入上下文感知模块,提高模型全局信息关联能力,并抑制无用背景突出人物特征。实验结果表明,相较于基准模型,YOLOv8Pose-DC的mAP@0.5提升3.1%,召回率提升4.2%。所设计的算法性能有较大提升,满足生产需要。

    Abstract:

    To address the issues of missed and false detections in existing dense crowd pose estimation algorithms, an improved YOLOv8sPose algorithm for dense crowd pose estimation, namely, YOLOv8Pose-Dense Crowd (YOLOv8Pose-DC), is proposed. First, a centralized intrinsic adjustment feature pyramid network is designed, which combines deformable attention mechanisms and coordinate attention-based spatial pyramid pooling fast (CASPPF) in a parallel manner, globally focusing and adjusting the pyramid network from top to bottom, thereby increasing the spatial weight of global representation within the network. This enables the improved algorithm to obtain comprehensive and distinctive feature representations. Second, a multi-scale dual detection head structure is proposed, reducing computational complexity while enhancing model detection efficiency. Furthermore, the DySample module is utilized to improve the upsampling efficiency of the model. Finally, a spatial context aware module (SCAM) is added to enhance the model's ability in associating global information and suppressing irrelevant background features, to highlight human characteristics. Compared to the baseline model, YOLOv8Pose-DC increases mAP@0.5 by 3.1% and recall rate by 4.2%. The designed algorithm significantly improves performance and fully meets production requirements.

    参考文献
    相似文献
    引证文献
引用本文

谷学静,郭志斌.层内调节特征金字塔的密集人群姿态估计算法[J].半导体光电,2024,45(6):931-938. GU Xuejing, GUO Zhibin. Dense Crowd Pose Estimation Algorithm for In-layer Adjustment Feature Pyramid[J].,2024,45(6):931-938.

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:2024-07-31
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2025-02-20
  • 出版日期:
文章二维码