基于关键点特征提取与融合的人体姿态检测模型

打开文本图片集

引用格式：，，，等.基于关键点特征提取与融合的人体姿态检测模型[J].现代电子技术，2026，49（7）：31-39.

关键词：姿态检测；注意力机制；特征提取；特征融合；空间金字塔卷积；CAFBlock网络中图分类号：TN911.73-34；TP391 文献标识码：A 文章编号：1004-373X（2026）07-0031-09

Human pose detection model based on keypoint feature extraction and fusion

LIUQiheng，HU Yongxiang，PAN Changning，HAN Longzhi （CollegeofRailway Transportation，HunanUniversityofTechnology，Zhuzhou412Ooo，China）

Abstract：Inviewofthelimitationsinkeypointfeatureextractionandfusioncapabilitiesoftheexisting modelsinvarious scenarios，thispaperproposesanimproved modelnamedYOLOvl1n-Pose.The modelreplaces theoriginalstructure Botleneck withaPKI（polyeelieption）bcktoanceeyointatureextractionAdditionallyultipledilationatespaialpad convolutionmoduleisesignedtoimproveflexibilityandexpressivenessuringfeatureextraction.Finall，theCAF（convolution atentionfusion）blocknetworkisintroducedtofurtherenhancethefusionofmulti-scalefeatures.Experimentalresults demonstratethattheprecisionrate，recallrate，mAP@0.5，andmAP@0.5：0.9oftheproposedalgorithmisincreasedby3.1%，2.9 % （20 3.5% ，and 1.2% ，respectively，on the COCO2017 dataset in comparison with thoseof the original model.In practical inference， thealgorithmsignificantlyreduceskeypointlocaizationerorsandmisseddetections，soitofersconsiderableaplcationvalue.

Keywords：pose detection；atention mechanism；featureextraction;feature fusion;spatial pyramid convolution;CAF block network

0 引言

人体姿态估计作为计算机视觉领域的重要研究方向，在智能监控、人机交互、虚拟现实等领域具有广泛的应用前景。（剩余12753字）