基于深度学习的人体行为识别检测综述

A Summary of Human Action Recognition and Detection Based on Deep Learning

人体行为识别检测是计算机视觉领域的研究热点,主要包括行为识别和行为检测两大部分。目前,对行为识别检测的综述主要聚焦在行为识别领域,对行为检测的关注度偏低。针对这一现状,聚焦行为识别和行为检测两个方面,分别综述了行为识别以及行为检测的各种方法,介绍了常用的数据集。首先从网络结构的角度重点论述了基于深度学习的行为识别方法;而后将行为检测划分为时序行为检测和时空行为检测,总结行为检测的各种算法;最后对各种算法的特点进行了总结分析,探索行为识别与行为检测的区别与联系,对当前研究面临的问题以及下一步的工作进行了总结和展望。

Human action recognition and detection is a hot research topic in the field of computer vision,which mainly includes action recognition and action detection.At present,summarizes in this filed are mainly skewed towards action recognition,and action detection is relatively underresearched.In view of this situation,focusing on both of them,the methods of action recognition and action detection were summarized and the commonly used datasets were introduced.Firstly,the action recognition methods based on deep learning were expounded from the perspective of network structure.Next,action detection was divided into temporal action detection and spatio-temporal action detection to summarize the algorithms of action detection.Finally,the characteristics of various algorithms and a discussion on differences and relations between action recognition and action detection were summarized,Also,the problems faced in current research and the expectations on future research were presented.