基于深度学习方法的人体行为识别综述  

A Review of Human Behavior Recognition Based on Deep Learning Methods

作　　者：张涵 冯佳红 ZHANG Han;FENG Jiahong(College of Science and Engineering,Jiaozuo Normal College,Jiaozuo 454000,Henan)

机构地区：[1]焦作师范高等专科学校理工学院,河南焦作454000

出　　处：《济源职业技术学院学报》2024年第4期62-69,共8页Journal of Jiyuan Vocational and Technical College

基　　金：焦作师范高等专科学校青年基金项目(2023-QJ-04)。

摘　　要：当前,人体行为识别技术广泛应用于公共安全监控、智能交通调度及个性化教学中。随着计算机视觉技术的飞跃,深度学习已成为行为识别领域的核心驱动力,显著提升了识别精度与效率,推动多领域智能化发展。为了梳理深度学习方法的发展脉络,对基于双流网络、3D卷积神经网络等框架中最具代表性的行为识别算法进行研究综述,分析各种算法的优缺点,同时对HMDB-51和UCF-101等几种主流数据集进行介绍,在UCF-101数据集上对经典算法进行性能对比,并探讨行为识别领域的研究趋势与未来潜在方向。Currently,human behavior recognition technology is widely used in public safety monitoring,intelligent traffic dispatch,and personalized teaching.With the leap of computer vision technology,deep learning has become the core driving force in the field of behavior recognition,significantly improving recognition accuracy and efficiency,and promoting the development of multi domain intelligence.In order to sort out the development of deep learning methods,this paper reviews the most representative behavior recognition algorithms based on frameworks such as two-stream networks and 3D convolutional neural networks,analyzes the advantages and disadvantages of various algorithms,introduces several mainstream data sets such as HMDB-51 and UCF-101,and compares the performance of classic algorithms on the UCF-101 dataset.Finally,the future research tendency and potential direction of human behavior recognition is prospected.

关 键 词：深度学习 行为识别 神经网络 数据集 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]