基于深度学习的二维人体姿态估计研究进展

Research Progress of Two-Dimensional Human Pose Estimation Based on Deep Learning

基于深度学习的二维人体姿态估计方法通过构建特定的神经网络架构,将提取的特征信息根据相应的特征融合方法进行信息关联处理,最终获得人体姿态估计结果,因其具有广泛的应用价值而受到研究人员的关注。从数据集基准、姿态估计方法和评测标准等方面,对近年来基于深度学习的二维人体姿态估计的诸多研究工作进行系统归纳与整理,将现有方法分为单人姿态估计方法与多人姿态估计方法,并分别从网络架构设计、输出特征表示和损失函数选取方面进行分析与总结。在此基础上,结合当前二维人体姿态估计所面临的挑战对其未来研究发展方向与应用前景进行展望。

The two-dimensional Human Pose Estimation(HPE)methods based on deep learning have attracted much attention for their application potential.The methods work by constructing a specific neural network architecture,and processing the extracted feature information based on the corresponding feature fusion method and information association strategy to obtain the human pose estimation result.This paper systematically summarizes the studies on two-dimensional human pose estimation based on deep learning in recent years,categorizing them into data set benchmarks,pose estimation methods and evaluation standards.The existing methods are divided into single-person pose estimation methods and multi-person pose estimation methods,and analyzed and summarized in terms of network architecture design,output feature representation and loss function selection.Finally,based on the current challenges,this paper discusses the development directions of future research and application prospects of two-dimensional human pose estimation.