A Guided Method for Improving the Video Human Action Classification in Convolutional Neural Networks

被引：0
|
作者：
Mao L. [1 ]
Chen S. [1 ]
Yang D. [1 ]
机构：
[1] College of Mechanical and Electronic Engineering, Dalian Minzu University, Dalian
来源：
Wuhan Daxue Xuebao (Xinxi Kexue Ban)/Geomatics and Information Science of Wuhan University | 2021年 / 46卷 / 08期
关键词：
3D convolutional neural networks; Dynamic information learning ability; Guided optimization; Temporal domain dynamic information understanding ability; Video human action classification;
D O I：
10.13203/j.whugis20190101
中图分类号：
学科分类号：
摘要：
Objectives: In order to improve the ability of convolutional neural networks (CNNs) of understanding temporal dynamic information, this paper proposes a dominant layer optimization module. Methods: The new module uses the dominant layer to guide and optimize the update gradient of convolutional layer weights, and assist the difference estimation with the maximum mean difference algorithm of a reproducing Hilbert space. Results: In continuous training, the network can improve the learning ability of temporal dynamic information, and the dynamic information similarity between the features learned by convolutional layer and the input data is also increased. Conclusions: This module enhances the performance of the CNNs model on video human action classification and achieves improvements to the network. © 2021, Editorial Board of Geomatics and Information Science of Wuhan University. All right reserved.
引用
收藏
页码：1241 / 1246
页数：5
相关论文
共 21 条
  • [1] Tran D, Bourdev L, Fergus R, Et al., Learning Spatiotemporal Features with 3D Convolutional Networks, 2015 IEEE International Conference on Computer Vision, (2015)
  • [2] Tran D, Wang H, Torresani L, Et al., A Closer Look at Spatiotemporal Convolutions for Action Recognition, IEEE Conference on Computer Vision and Pattern Recognition, (2018)
  • [3] Pei Songwen, Yang Baoguo, Gu Chunhua, Research on Video Stream Classification Using 3D ConvNet Ensemble Fusion Model, Journal of Chinese Computer Systems, 39, 10, pp. 2266-2270, (2018)
  • [4] Wu Peiliang, Yang Xiao, Mao Bingyi, Et al., A Perspective⁃Independent Method for Behavior Recognition in Depth Video via Temporal⁃Spatial Correlating, Journal of Electronics and Information Technology, 41, 4, pp. 904-910, (2019)
  • [5] Simonyan K, Zisserman A., Two⁃Stream Convolutional Networks for Action Recognition in Videos, Advances in Neural Information Processing Systems, (2014)
  • [6] Sevilla-Lara L, Liao Y, Guney F, Et al., On the Integration of Optical Flow and Action Recognition, German Conference on Pattern Recognition, (2018)
  • [7] Huang D A, Ramanathan V, Mahajan D, Et al., What Makes a Video a Video: Analyzing Temporal Information in Video Understanding Models and Datasets, IEEE Conference on Computer Vision and Pattern Recognition, (2018)
  • [8] Xiong Hanjiang, Zheng Xianwei, Ding Youli, Et al., Semantic Segmentation of Indoor 3D Point Cloud Model Based on 2D⁃3D Semantic Transfer, Geomatics and Information Science of Wuhan University, 43, 12, pp. 2303-2309, (2018)
  • [9] Luo Z, Hsieh J T, Jiang L, Et al., Graph Distillation for Action Detection with Privileged Modalities, European Conference on Computer Vision, (2018)
  • [10] Wang X, Girshick R, Gupta A, Et al., Non⁃local Neural Networks, IEEE Conference on Computer Vision and Pattern Recognition, (2018)
123