基于人体关节点视觉特征的姿态识别方法研究进展  

Study Progress in Posture Recognition Method Based on Visual Features of Human Joints

作　　者：梅健 王甜甜 黄凯 邹青宇 MEI Jian;WANG Tian-tian;HUANG Kai;ZOU Qing-yu(College of Information and Control Engineering,Jilin Institute Chemical Technology;College of Electrical and Information Technology,Beihua University)

机构地区：[1]吉林化工学院信息与控制工程学院 [2]北华大学电气与信息工程学院

出　　处：《化工自动化及仪表》2023年第2期119-124,共6页Control and Instruments in Chemical Industry

基　　金：中央引导地方科技发展资金吉林省基础研究专项(YDZJ202101ZYTS170);教育部产学合作协同育人项目(202102181046);2020年度吉林省高等教育教学研究课题;国家大学生创新创业训练计划项目(202210201055);吉林省职业教育与成人教育教学改革研究课题(PX-521509);吉林市科技创新发展计划项目(20210103098)。

摘　　要：随着深度学习和人工智能的发展,相关算法在人体姿态识别领域得到了广泛应用。基于人体关节点智能化算法的姿态识别得到了利用和优化,现有的研究成果不同程度涉及各类算法,亟待系统性总结和分类。综述近年来基于视觉特征的姿态识别领域的研究成果,将这些方法分为两类:单人姿态识别和多人姿态识别。分析各方法的原理和突出进步,并介绍基于OpenPose方法的多人姿态研究,指出了未来研究的趋势与焦点。With the development of deep learning and artificial intelligence, relevant algorithms have been widely applied in human postures recognition and the posture recognition based on the intelligent algorithms for human joints has been optimized. Existing research results involve all kinds of algorithms at various degrees, wait for being systematically summarized and classified. A large number of recent research achievements in the posture recognition were reviewed and divided into single-person posture recognition and multiperson posture identification, including having their principles and outstanding progress analyzed, and their future research trends and hot spots were pointed out based on describing the OpenPose-based study of the multi-person postures.

关 键 词：人工智能 深度学习 多人姿态识别 智能化算法 OpenPose 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]