摘要
主要讲述人体行为识别的基础流程,归纳了人体行为识别常用的数据集,总结了时域分割的发展现状和常用的方法,讲解了人体行为识别比较经典的方法,并归纳了人体行为识别最新、最热的深度学习方法。引入了动作分割,再结合行为识别,能够实现连续的人体行为识别,使得行为识别适用于实际场景,而不再是对经过人工剪辑好的单个视频进行识别,这在实际应用中意义重大。
This paper focused on action recognition and included data sets and motion segmentation.It mainly described the basic flow of human action recognition.And it summarized the commonly used data sets of human action recognition.Then it summarized the development status and common methods of time domain segmentation.Next it explained the classic algorithms of human action recognition.At last,it summarized the the-state-of-art deep learning methods of human action recognition.The introduction of action recognition combines with action segmentation made the action recognition applicable to the actual scene,which could achieves continuous recognition of human action.Meanwhile it was no longer recognize a single video that has been manually edited.This has very important reference value in practical applications.
作者
陈煜平
邱卫根
Chen Yuping;Qiu Weigen(School of Computers, Guangdong University of Technology, Guangzhou 510006, China)
出处
《计算机应用研究》
CSCD
北大核心
2019年第7期1927-1934,共8页
Application Research of Computers
基金
国家自然科学基金资助项目(61572142)
广东省科技计划资助项目(14ZK0180)
关键词
人体行为识别
数据集
动作分割
深度学习
双流网络
human action recognition
data set
motion segmentation
deep learning
two-stream network
