Open AccessComparative analysis of 3D convolutional and LSTM neural networks in the action recognition task by video data
Author(s) -
Ruslan J. Portsev,
А. В. Макаренко
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1864/1/012015
Subject(s) - computer science , convolutional neural network , generalization , task (project management) , artificial intelligence , action recognition , pattern recognition (psychology) , artificial neural network , action (physics) , neocognitron , machine learning , time delay neural network , class (philosophy) , mathematics , engineering , mathematical analysis , physics , quantum mechanics , systems engineering
In the present paper a comparative analysis of two architectural neural network approaches (based on 3D convolutional and LSTM) in the recognition of actions on video is made. The problem was being solved on 10 behavior classes separated from the UCF50 dataset. The original neural network architectures were developed and pre-trained. It was found that the network based on 3D convolutions has better generalization ability and is more stable in the training.