摘要
提出了一种基于三维卷积稠密网络的视频行为识别算法,利用三维卷积进行卷积操作,获取视频中人体行为的特征,并基于DenseNet的连接方式进行网络层级连接,得到高维特征,从而构建三维卷积稠密Dense-3D网络,以进行视频中的人体行为识别。分别在数据集KTH和UCF-101上进行测试,实验结果均表明所构建网络结构在视频行为识别任务中有着较好的识别效果。
An action recognition algorithm based on the three-dimensional convolutional network is proposed.Three-dimensional convolution is used to obtain the characteristics of human action.The network hierarchy based on DenseNet’s connection method is applied to obtain high-dimensional features for constructing a three-dimensional convolution dense network,named as Dense-3D.The experimental results on the datasets KTH and UCF-101 show that the generated network presents a good recognition effect.
作者
李刚
刘新
顾广华
LI Gang;LIU Xin;GU Guanghua(School of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;Hebei Key Laboratory of Information Transmission and Signal Processing,Yanshan University,Qinhuangdao,Hebei 066004,China)
出处
《中国科技论文》
CAS
北大核心
2018年第14期1634-1638,1663,共6页
China Sciencepaper
基金
河北省自然科学基金资助项目(F2017203169)
河北省高等学校科学研究重点项目(ZD2017080)
河北省留学回国人员科技活动项目(CL201621)
