基于循环神经网络的人体动作在线识别方法
Research of Human Action Online Recognition Based on Recurrent Neural Network
DOI: 10.12677/CSA.2024.141013, PDF,   
作者: 李高丰：洛阳师范学院，物理与电子信息学院，河南 洛阳
关键词: 动作识别；姿态检测；循环神经网络；注意力；Action Recognition； Pose Estimator； Recurrent Neural Network； Attention
摘要: 由于人体运动的非刚性以及表观特征的多样性，现实场景中对人体动作的准确识别较难。本文提出一种基于循环神经网络的人体动作在线识别方法，把人体关节点作为主要特征，利用深度学习进行动作识别建模。采用循环神经网络学习连续动作的时序关联信息，引入随机权重共享的注意力机制提高训练准确率，避免过拟合现象，实现对人体动作的在线识别。通过UCF数据集进行训练和测试，本文方法达到较高的准确率和稳定性，表明了基于循环神经网络的动作识别模型对现实场景人体动作的在线识别是有效的。
Abstract: Due to the non-rigidity of human motion and the diversity of apparent features, it is difficult to ac-curately identify human action in real scenes. In this paper, an online recognition method of human action based on recurrent neural network is proposed. The human joint points are used as the main features. The deep learning is used to model the action recognition. The recurrent neural network learns the temporal correlation of continuous actions and realizes online recognition of human ac-tions, in which the attention mechanism is introduced to improve the training accurate rate and decrease over fitting. Through the training and testing of UCF dataset, the proposed method achieves high accuracy and stability, which shows that the action recognition model based on re-current neural network is effective for online recognition of human motion in real scenes.
文章引用：李高丰. 基于循环神经网络的人体动作在线识别方法[J]. 计算机科学与应用, 2024, 14(1): 113-122. https://doi.org/10.12677/CSA.2024.141013

参考文献

[1] 罗会兰, 童康, 孔繁胜. 基于深度学习的视频中人体动作识别进展综述[J]. 电子学报, 2019, 47(5):1162-1173.
[2] 邵志文, 周勇, 谭鑫. 基于视觉的人体动作识别综述基于深度学习的表情动作单元识别综述[J]. 电子学报, 2022, 50(8): 2003-2017.
[3] 包本刚. 融合多特征的目标检测与跟踪方法[J]. 电子测量与仪器学报, 2019, 33(9): 93-99.
[4] 秦磊, 胡琼, 黄庆明, 等. 基于特征点轨迹的动作识别[J]. 计算机学报, 2014, 37(6): 1281-1288.
[5] Mekruksavanich, S. and Jitpattanakul, A. (2022) Multimodal Wearable Sensing for Sport-Related Ac-tivity Recognition Using Deep Learning Networks. Journal of Advances in Information Technology, 13, 132-138. [Google Scholar] [CrossRef] 
[6] Sevilla-Lara, L., Liao, Y.Y., Guney, F., et al. (2017) On the Inte-gration of Optical Flow and Action Recognition. In: Brox, T., Bruhn, A. and Fritz, M., eds., Pattern Recognition. GCPR 2018. Springer, Cham.
[7] 朱煜, 赵江坤, 王逸宁, 郑兵兵. 基于深度学习的人体行为识别算法综述[J]. 自动化学报, 2016, 42(6): 848-857.
[8] Simonyan, K. and Zisserman, A. (2014) Two-Stream Convolutional Networks for Action Recognition in Videos.
https://arxiv.org/abs/1406.2199
[9] 曹晋其, 蒋兴浩, 孙锬锋. 基于训练图CNN特征的视频人体动作识别算法[J]. 计算机工程, 2017, 43(11): 234-238.
[10] Zha, S., Luisier, F. and Rews, W., et al. (2015) Exploiting Im-age-Trained CNN Architectures for Unconstrained Video Classification.
https://arxiv.org/abs/1503.04144
[11] Krizhevsky, A., Sutskever, I. and Hinton, G. (2012) ImageNet Classification with Deep Convolutional Neural Networks. Conference and Workshop on Neural Information Processing Systems, Lake Tahoe, Nevada, 3-8 December 2012.
[12] Simonyan, K. and Zisserman, A. (2015) Very Deep Convolutional Networks for Large-Scale Image Recognition.
https://arxiv.org/abs/1409.1556
[13] Szegedy, C., Liu, W., Jia, Y.Q., et al. (2014) Going Deeper with Convolu-tions. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 7-12 June 2015. [Google Scholar] [CrossRef] 
[14] Huang, G., Liu, Z., Maaten, L.V.D., et al. (2017) Densely Con-nected Convolutional Networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21-26 July 2017. [Google Scholar] [CrossRef] 
[15] Wei, L.R., Yue, J., Zhu, H., et al. (2019) Human Action Recognition Method Based on Deep Neural Network. Journal of University of Jinan (Science and Technology), 33, 215-223.