融合注意力机制的人体关键姿态估计  

Human key pose estimation based on attention mechanism

作　　者：冯霞[1,2] 薛晶霞 刘才华 FENG Xia;XUE Jing-xia;LIU Cai-hua(School of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China;The Key Laboratory of Smart Airport Theory and System,Civil Aviation Administration of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300 [2]中国民用航空局智慧机场理论与系统重点实验室,天津300300

出　　处：《计算机工程与设计》2023年第9期2754-2760,共7页Computer Engineering and Design

基　　金：天津市教委科研计划基金项目(2021KJ037);中央高校基本科研业务费基金项目(3122021052);民航航空公司人工智能重点实验室自主课题基金项目(CZAILAB-COO-KJAI20001)。

摘　　要：为提高与特定动作相关的目标关键姿态识别准确率,提出融合注意力机制的人体关键姿态估计方法。针对特征提取冗余和通道信息损失问题,在网络不同阶段融入注意力机制,帮助模型从空间和通道维度提取对任务更加重要的特征;考虑与特定动作相关的目标关键姿态误识别代价更大,设计一种代价敏感损失函数,提升模型对目标关键姿态的识别性能。实验结果表明,所提方法在COCO数据集和CargoSorting数据集上对目标关键姿态识别精度分别提高了1.4%和1.5%。To improve the accuracy of target key pose recognition related to specific actions,a human key pose estimation method integrating attention mechanism was proposed.Aiming at the redundancy of feature extraction and the loss of channel information,attention mechanism was integrated into different stages of the network to help the model extract more important features from the dimensions of space and channel.Considering the higher cost of target key pose recognition related to specific actions,a cost sensitive loss function was designed to improve the performance of target key pose recognition.Experimental results show that the proposed method improves the accuracy of target key pose recognition by 1.4%and 1.5%on COCO dataset and CargoSorting dataset respectively.

关 键 词：人体姿态估计 目标关键姿态 通道信息损失 特征冗余 空间注意力 通道注意力 代价敏感 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]