基于堆叠沙漏网络改进的多人姿态估计  

Improved multi-person pose estimation based on stacked hourglass networks

作　　者：梁鸿[1] 任文静 张千[1] 李传秀[1] LIANG Hong;REN Wen-jing;ZHANG Qian;LI Chuan-xiu(College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《计算机工程与设计》2022年第2期502-509,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61673396);2018年中央高校基本科研业务费基金项目(18CX02138A)。

摘　　要：为解决多人姿态估计中小尺度关键点(手腕、脚踝等)定位准确率低的问题,采用自顶向上的方式,结合先进的人体目标检测模型YOLOv3,提出一种基于堆叠沙漏网络改进的多人姿态估计方法。在沙漏网络中融入通道混洗模块,加强不同尺度下多层特征之间的跨通道信息交流,提高被遮挡关键点的识别效果;使用注意力机制对沙漏网络原有的残差模块进行特征增强,抑制无用特征并提升有用特征,提高小尺度关键点的识别率。实验结果表明,在MPII数据集上的总体PCK@0.5达到了88.6%,在MSCOCO数据集上的AP@0.75相比原始网络提升了4.6%,验证了所提方法的有效性。To deal with the problem that the low positioning accuracy of small-scale key points(wrist,ankle,etc.)in multi-person pose estimation,a modified multi-person pose estimation algorithm based on stacked hourglass networks was proposed using the top-down method and the advanced human target detection model YOLOv3.The channel shuffling module was integrated into the stacked hourglass networks to enhance the cross-channel information exchange between multi-layer features under diffe-rent scales to improve the recognition effect of occluded key points.Attention mechanism was used to enhance the original residual module of hourglass networks to suppress useless features and enhance useful features,so as to improve the recognition rate of small-scale key points.Experimental results show that the overall PCK@0.5 reaches 88.6%on the MPII datasets.Compared with original networks,the AP@0.75 is improved by 4.6%on the MSCOCO datasets,and the validity of the proposed method is verified.

关 键 词：堆叠沙漏网络 人体姿态估计 多尺度特征融合 注意力机制 人体目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]