重庆科技大学学报（自然科学版）2025，Vol.27Issue(1)：83-89,7.DOI:10.19406/j.issn.2097-4531.2025.01.011
基于改进YOLOv8n-Pose的人体关键点检测算法
Human Key Point Detection Algorithm Based on Improved YOLOv8n-Pose
摘要
Abstract
Aiming at the limitations of computational resource and storage space,which the YOLOv8n-Pose model faces when deployed on edge computing devices,this paper proposes an improved human key point detection algo-rithm based on the improved YOLOv8n-Pose model.A context-guided feature fusion model is designed to improve the expression of CGFF model's feature and detection accuracy.Meanwhile,the depthwise convolution is intro-duced to improve the head part of the model,significantly reducing the number of parameters and calculations.The experimental results of the COCO-Pose dataset show that the optimized YOLOv8n-Pose model reduces the overall network parameters by 19.5%and the amount of calculation by 32.6%while ensuring the improvement of detection accuracy,which improves its applicability in edge computing scenarios.关键词
YOLOv8n-Pose算法/深度卷积/上下文导引特征融合/姿态估计Key words
YOLOv8n-Pose algorithm/depthwise convolution/context-guided feature fusion/pose estimation分类
信息技术与安全科学引用本文复制引用
雷亮,陈毅,刘学涵,赵锦,陈小庆,周华勇,徐山雯..基于改进YOLOv8n-Pose的人体关键点检测算法[J].重庆科技大学学报（自然科学版）,2025,27(1):83-89,7.基金项目
重庆市教委科学技术研究项目"基于应急决策的校园社区大数据平台研究与应用"(KJQN202203314),"基于YOLO-Pose姿态特征检测融合时序的独居老人跌倒识别算法研究"(KJQN202303305),"基于深度学习的学生课堂行为检测方法研究"(KJQN202403322) （KJQN202203314）
2021年重庆市属本科高校与中科院所属院所合作项目"工业互联网内生安全关键技术研究与协同创新"(HZ2021015) （HZ2021015）
