首页|期刊导航|生物医学工程研究|基于改进YOLOv8n-Pose的人体关键点检测模型

基于改进YOLOv8n-Pose的人体关键点检测模型OA

Human keypoint detection model based on improved YOLOv8n-Pose

中文摘要英文摘要

针对多人、遮挡等复杂场景下,人体关键点识别准确率低的问题,本研究基于 YOLOv8n-Pose 提出了改进模型.首先,引入 DCNv3 模块,在增强网络准确率的同时,降低网络的计算复杂度;其次,引入空间和通道协同注意力(SCSA)机制,提高网络的特征提取能力;最后,引入 EIoU 损失函数,加快边界框检测的收敛速度.改进模型在 OC_Human 数据集上的测试结果表明,相比于原始模型,精度提升 2.1%,mAP50 提升 3.0%,浮点运算量降低 0.7%.在复杂场景下,本研究模型能准确、有效地识别人体关键点,满足多人场景下,人体关键点检测任务的需求.

To address the problem of low accuracy in human keypoint recognition under complex scenarios such as multiple people and occlusion,we proposed an improved model based on YOLOv8n-Pose.Firstly,the DCNv3 module was introduced to enhance the accuracy of the network while reducing its computational complexity.Then,the spatial and channel synergistic attention(SCSA)mech-anism was introduced to enhance the network's feature extraction capability.Finally,the EIoU loss function was used to accelerate the convergence speed of bounding box detection.The test results of the improved model on the OC_Human dataset showed that,compared with the original model,the accuracy was increased by 2.1%,mAP50 was improved by 3.0%,and floating point operations was re-duced by 0.7%.The proposed model can accurately and effectively identify human keypoint detection in complex scenarios,and meet the requirements for human keypoint detection tasks in multi-person scenarios.

李宇皓;仇大伟;郭振

山东中医药大学 医学信息工程学院,济南 250014山东中医药大学 医学信息工程学院,济南 250014山东中医药大学 医学信息工程学院,济南 250014

医药卫生

姿态估计人体关键点图像处理检测注意力机制

Pose estimationHuman keypointImage processingDetectionAttention mechanism

《生物医学工程研究》 2026 (4)

304-310,7

10.19529/j.cnki.1672-6278.2026.04.06

评论