基于深度学习的动作识别方法简述

A Survey of Action Recognition based on Deep Learning

基于深度学习的动作识别是计算机视觉、机器学习等多学科的交叉研究课题,在体育比赛智能化、医疗看护、安防等领域有着广阔应用前景。本文简述了三类动作识别深度学习算法的近年研究现状,即CNN-LSTM框架、3D卷积框架以及双流网络框架,并对国际上用于动作识别的常用数据库进行简单介绍;最后进行了总结和展望。

Action recognition based on deep learning is a cross-disciplinary research topic in many fields such as computer vision and machine learning.It has broad application prospects in the fields of sports competition intelligence,medical care,and security.This paper briefly describes the current research status of three action recognition deep learning frameworks,namely CNN-LSTM framework,3D convolution framework and two-stream network framework,and briefly introduces the commonly used database for motion recognition in the world.Finally,the paper summarizes the two methods and gives the prospect.