摘要
随着深度学习的发展,相关的模型与方法被应用到单人姿态估计研究领域中并取得了较大提升,一举超越了传统基于手工设计特征的单人姿态估计方法.基于深度学习方法的单人姿态估计从各个方面取得了突破.本文总结了基于深度学习方法的单人姿态估计这一领域一系列研究成果,将这些方法分为三类:基于坐标回归的方法、基于热力图回归的方法以及使用热力图表示的基于坐标回归的方法.着重介绍了基于热力图回归这一主流的方法,分析并对比了各类方法的优势与不足.给出了单人姿态估计常用数据集的对比,并对基于深度学习方法的单人姿态估计研究进行展望,指出了未来研究的趋势与热点.
With the development of deep learning,related models and methods are introduced to the field of single-person pose estimation and result in a significant improvement,surpassing those methods using hand-crafted features.The deep learning based single-person pose estimation has made breakthroughs in various aspects.This paper summarizes the recent research of deep learning based single-person pose estimation,and divides these methods into three categories:coordinate based regression methods,heatmap based regression methods,and coordinate based regression methods using heatmap representation.This paper focuses on the main stream methods based on heatmap regression,and analyzes and compares the advantages and disadvantages of various methods.A comparison of commonly used datasets is given.In the end of this paper,the future direction and research trends of single-person pose estimation are discussed.
作者
张锋
叶茂
曾凡玉
ZHANG Feng;YE Mao;ZENG Fan-yu(School of Computer Science and Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第7期1501-1507,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金-联合基金项目(U181320052)资助
国家自然科学基金面上项目(6177020680)资助
国家重点研究计划项目(2018YFC0831801)资助
四川省重点研发项目(17ZDYF3184)资助。
关键词
单人姿态估计
身体部件检测
关节点定位
坐标回归
热力图回归
single-person pose estimation
body part detection
key point localization
coordinate regression
heatmap regression
