1 School of Mechanical Engineering, Shenyang Jianzhu University, Shenyang 110168, China
2 National Key Laboratory of Digital Steel, Northeastern University, Shenyang 110819, China
Show less
文章历史+
Received
Published
Revised
2025-09-16
2026-05-28
2025-11-27
Online Date
2026-08-07
PDF (7075K)
摘要
在工业自动化与智能化趋势下,行为识别技术通过“实时监测-预警-干预”的全流程支持,将工厂安全生产从“事后补救”转向“事前预防”,既保障了工人生命安全,也为企业降低了事故损失、提升了生产稳定性。针对工厂实际复杂场景中物体遮挡人体导致的行为识别精度下降问题,提出一种基于特征自校准和特征精炼的遮挡人体姿态估计算法。首先,利用特征自校准单元(feature self-calibration unit, FSCU),自适应建模学习空间和通道之间的依赖关系,在产生全局感受野的同时通过通道信息减少无关区域的影响,获取更加具有区分度的关节点特征表示。之后,设计一种多粒度特征精炼单元(multi-granularity feature refinement unit, MGFRU),在空间和通道维度均保持较高分辨率,聚焦感知不同区域的关键信息,对骨干网络提取的不同粒度特征进行筛选与精炼,突出与关节点相关的关键特征。通过与主流模型分析对比,在MPII公开数据集上的总体平均准确率达到90.8%。优于所列的其他主流模型。在模拟不同遮挡程度的自建工厂场景数据集上,本模型展现了更强的鲁棒性,在实际工程中提高了生产安全性与预防可靠性。
Abstract
Amid the trends of industrial automation and intelligent development, human action recognition technology provides full-process support encompassing “real-time monitoring-early warning-intervention”. It transforms factory work safety management from “post-accident remedy” approach to “pre-accident prevention”, which not only safeguards workers’ lives but also helps enterprises reduce accident losses and improve production stability. To address the decline in action recognition accuracy caused by object occlusion of the human body in complex real-world factory scenarios, an occluded human pose estimation algorithm based on feature self-calibration and feature refinement was proposed. Firstly, a feature self-calibration unit (FSCU) was utilized to adaptively model and learn the dependency relationships between spatial and channel dimensions. While generating a global receptive field, it reduced the impact of irrelevant regions through channel information, thereby obtaining more discriminative feature representations of joint points. Subsequently, a multi-granularity feature refinement unit (MGFRU) was designed, which maintained high resolution in both spatial and channel dimensions and focuses on perceiving key information in different regions. This unit filters and refines multi-granularity features extracted by the backbone network, highlighting key features related to joint points. The model proposed was analyzed and compared with mainstream models, the proposed model achieves an overall mean accuracy of 90.8% on the MPII public dataset, outperforming all other listed state-of-the-art models. On the self-constructed factory scenario dataset that simulates different occlusion levels, this model exhibits stronger robustness, thus improving production safety and preventive reliability in practical engineering applications.
ZHANG Tian, WU Xiao-lin, LIU Ke-pu, TIAN Yong, WANG Bing-xing.
Occluded Human Pose Estimation Model Based on Feature Self-calibration and Refinement[J].
Science Technology and Engineering, 2026, 26(15): 6487-6495 DOI:10.12404/j.issn.1671-1815.2506661
受以往研究的启发,为了更深度挖掘多分辨率特征信息并有效融合,以提高遮挡场景下的姿态估计准确度,提出一种基于特征自校准和特征精炼的遮挡人体姿态估计算法SCR-HRNet(self-calibration and refinement based high-resolution network)。该网络在HRNet结构基础上引入特征自校准单元(feature self-calibration unit, FSCU) 自适应地建立长距离空间和通道信息之间的相关性,产生全局感受野的同时通过通道信息减少无关区域特征影响。同时设计多粒度特征精炼单元 (multi-granularity feature refinement unit, MGFRU),旨在保持高分辨率的同时捕捉全局与局部特征关系,进行特征提纯,有效融合高分辨率空间信息与低分辨率语义信息,从而提高网络的抗干扰能力,改善遮挡场景下的姿态估计效果。
在MPII数据集进行姿态估计模型改进效果的验证。MPII数据集是一个主要针对人体姿态估计任务的数据集,它的主要评估指标是PCKh分数(percentage of correct keypoints with head normalization),指预测正确关键点的比例,将预测得到的关键点与标注关键点之间的归一化距离计算出来,若小于设定阈值α则认为预测正确,本文中选择的评估标准为PCKh@0.5,即设定的阈值α=0.5,在评估过程中使用多个关键点进行评估。
4.2 实验配置与评价指标
为了与其他方法进行公平比较,整个训练过程按照原始算法训练策略,固定人体检测框的长宽比为4∶3,MPII数据集上输入大小剪裁为256×256,为了防止过拟合,进行数据增强操作,包括随机旋转、随机缩放以及随机翻转,实验Adam优化器,批量大小为32,初始学习率设置为1×10-3,共训练210轮,在第170轮和第200轮分别将为1×10-4和1×10-5。在测试过程中遵循HRNet相同的检测方式,在MPII数据集中使用官方原始提供的人体位置框,模型在一台NVIDIA RTX 4060 Ti GPU上训练。
YangShilin, GaoZhijun, BuChunguang, et al. Human joint point detection method based on rescue robots[J]. Science Technology and Engineering, 2024, 24(17): 7214-7220.
WangCailing, YanJingjing, ZhangZhidong. Review on human action recognition methods based on multimodal data[J]. Computer Engineering and Applications, 2024, 60(9): 1-18.
WuDonghui, XuJing, ChenJibin, et al. Human action recognition based on fusion of attention mechanism and CNN-LSTM[J]. Science Technology and Engineering, 2023, 23(2): 681-689.
WuNing, WangPeng, LiXiaoyan, et al. Lightweight human pose estimation based on adaptive feature sensing[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(8): 1107-1117.
JiangYuehan, ChenJunjie, LiHongjun. Review of human action recognition based on skeletal graph neural networks[J]. Computer Engineering and Applications, 2025, 61(3): 34-47.
WangXiaopeng, ShiHuan. Fall detection algorithm based on the improved YOLOv8 combined with key points[J]. Journal of Xidian University, 2024, 51(5): 149-164.
[13]
LiY, ZhangS, WangX, et al. TokenPose: learning keypoint tokens for human pose estimation[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. New York: IEEE, 2023: 11245.
[14]
ChenT, ZhouW, LiH. Graph-based human pose estimation with occlusion-aware feature learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(8): 3789-3802.
[15]
WangX, LiD, ChenK, et al. A real-time occlusion-aware pose estimation system for safety monitoring in steel plants[C]// IEEE International Conference on Robotics and Automation. New York: IEEE, 2025: 1-8.
YuBaoling, YuSongkun, SunYaoran, et al. Human body joint nodes detection based on DeepPose and Faster RCNN[J]. Journal of University of Chinese Academy of Sciences, 2020, 37(6): 828-834.
[18]
HuangH, LiD, ZhangZ, et al. Adversarially occluded samples for person reidentification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 5098-5107.
[19]
WeiS E, RamakrishnaV, KanadeT, et al. Convolutional pose machines[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 4724-4732.
[20]
NewellA, YangK, DengJ. Stacked Hourglass networks for human pose estimation[C]// Computer Vision-ECCV 2016: 14th European Conference. Berlin: Springer International Publishing, 2016: 483-499.
[21]
ChenY, WangZ, PengY, et al. Cascaded pyramid network for multi-person pose estimation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 7103-7112.
XieYu, BaoZiqun, ZhangNa, et al. Object detection algorithm based on feature enhancement and deep fusion[J]. Journal of Zhejiang University (Engineering Science), 2022, 56(12): 2403-2415.
NiuYue, WangAnnan, WuShengxi. Pose estimation based on attention module and CPN[J]. Journal of East China University of Science and Technology, 2023, 49(5): 724-734.
[26]
SunK, XiaoB, LiuD, et al. Deep high-resolution representation learning for human pose estimation[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2019: 5693-5703.
[27]
MoonG, ChangJ Y, LeeK M. PoseFix: model-agnostic general human pose refinement network[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2019: 7773-7781.
[28]
PishchulinL, InsafutdinovE, TangS, et al. Deepcut: joint subset partition and labeling for multi person pose estimation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 4929-4937.
[29]
CaoZ, HidalgoG, SimonT, et al. OpenPose: realtime multi-person 2D pose estimation using part affinity fields[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(1): 172-186.
[30]
WangJ, LongX, GaoY, et al. Graph-PCNN: Two stage human pose estimation with graph pose refinement[C]// The 16th European Conference on Computer Vision. Berlin: Springer International Publishing, 2020: 492-508.
[31]
TanM, PangR, LeQ V. EfficientDet: scalable and efficient object detection[C]// The IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 10781-10790.
[32]
ZhaoS G, YiW, ShiJ J, et al. An innovative crack detection algorithm based on efficient feature fusion and progressive transfer learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(12): 21469-21483.
[33]
XuJ, LiuW, XingW, et al. MSPENet: multi-scale adaptive fusion and position enhancement network for human pose estimation[J]. The Visual Computer, 2023, 39(5): 2005-2019.
[34]
XiaoB, WuH, WeiY. Simple baselines for human pose estimation and tracking[C]// Proceedings of the European Conference on Computer Vision (ECCV). Berlin: Spring, 2018: 466-481.
[35]
SuZ, YeM, ZhangG, et al. Cascade feature aggregation for human pose estimation[J]. arXiv Preprint, arXiv: 1902. 07837, 2019.
[36]
HuoZ, JinH, QiaoY, et al. Deep high-resolution network with double attention residual blocks for human pose estimation[J]. IEEE Access, 2020, 8: 224947-224957.
[37]
ZhangF, ZhuX, DaiH, et al. Distribution-aware coordinate representation for human pose estimation[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 7093-7102.