基于全局一致性网络的参数化人体网格重建  

Parametric Human Body Mesh Reconstruction Based on Global Consistency Network

作　　者：鲍文霞[1] 田如震 王年[1] 陈和木[2] 杨先军[3] BAO Wenxia;TIAN Ruzhen;WANG Nian;CHEN Hemu;YANG Xianjun(School of Electronic and Information Engineering,Anhui University,Hefei 230601,Anhui,China;The First Affiliated Hospital of Anhui Medical University,Hefei 230022,Anhui,China;Hefei Institute of Physical Sciences,Chinese Academy of Sciences,Hefei 230031,Anhui,China)

机构地区：[1]安徽大学电子信息工程学院,安徽合肥230601 [2]安徽医科大学第一附属医院,安徽合肥230022 [3]中国科学院合肥物质科学研究院,安徽合肥230031

出　　处：《华南理工大学学报（自然科学版）》2024年第7期19-28,共10页Journal of South China University of Technology(Natural Science Edition)

基　　金：国家重点研发计划项目(2020YFF0303803);安徽省重点研究与开发计划资助项目(2022k07020006);安徽高校自然科学研究资助项目(KJ2021ZD0004,2022ah051160)。

摘　　要：人体网格重建(HMR)在人机交互和虚拟/增强现实等领域有广泛应用。为了进一步提高基于图像的人体网格重建中人体姿势和形状估计的精度,提出了基于混合逆运动学的全局一致性深度卷积神经网络,用于参数化人体网格重建(GloCoNet)。为了增强网络的全局一致性和全局上的长程依赖,该网络在特征提取网络基础上,设计了全局一致性增强器(GCB)模块,它能够增强模型对全局信息的感知能力和表达能力,并且使模型能够自适应地调整不同通道和空间位置的特征图权重。然后引入了多头注意力机制(MHSA)来捕获模型全局上的长程依赖,它可以帮助模型在处理长期依赖时更好地捕捉到关键的关系和模式,并建模全局上下文信息,从而更好的丰富特征子空间的多样性。同时,该网络采用混合逆运动学的方法弥合人体网格估计和3D人体关节点估计之间的差距,最终提升人体3D姿势和形状估计的准确度。实验结果表明,GloCoNet模型在公开的Human 3.6 M数据集上以平均每关节51.3 mm的位置误差(MPJPE)显著优于先前的主流方法。Human body mesh reconstruction(HMR)has wide applications in human-computer interaction,virtual/augmented reality,and other fields.In order to further improve the accuracy of human body pose and shape estimation in image-based human body mesh reconstruction,this study proposed a parametric human body mesh reconstruction network based on hybrid inverse kinematics and global consistency deep convolutional neural network,called GloCoNet.To enhance the network’s global consistency and long-range dependencies,a Global Consistency Booster(GCB)module was designed on top of the feature extraction network.It can enhance the model’s perception and expression capabilities of global information,and allow the model to adaptively adjust the feature map weights of different channels and spatial positions.Furthermore,a multi-head attention mechanism was introduced to capture the model’s long-range dependencies globally,helping the model better capture key relationships and patterns when dealing with long-term dependencies,and modeling global contextual information to enrich the diversity of feature subspaces.Meanwhile,the network adopts a hybrid inverse kinematics approach to bridge the gap between human body mesh estimation and 3D human joint estimation,ultimately improving the accuracy of human 3D pose and shape estimation.Experimental results show that the GloCoNet model significantly outperforms previous mainstream methods with an average per joint position error of 51.3 mm on the publicly available Human3.6M dataset.

关 键 词：人体网格重建 全局一致性 混合逆运动学 人体参数 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]