Introducing a New Architecture of Deep Belief Networks for Action Recognition in Videos

Articles in Press

Document Type : Research Paper

Authors

1 PhD Student of Computer Engineering, Faculty of Electrical and Computer Engineering, University of Kashan , Kashan, Iran

2 Dept. of Electrical and Computer Engineering, University of Kashan, Kashan, Iran

Abstract

Deep learning has been widely used to solve problems related to the analysis of complex and voluminous data such as videos. One of the processing tasks performed on videos is human action recognition, which has important applications in the field of automatic surveillance, human-computer interaction, and the study of elderly behavior. Deep belief networks (DBNs) have been particularly attractive among different types of deep networks because of their special features, especially their ability to converge faster than other methods and the identical structure of their layers. However, the power of basic DBNs in processing complex data that are also time-dependent is worth considering.
In this paper, a new recurrent method based on DBNs is proposed. In the proposed method, the ability to process and interpret two-dimensional video frames and understand the concept of time through recursive implementation is added to DBNs. This method is capable of understanding short-term temporal concepts using restricted Boltzmann machines and long-term temporal concepts based on recursive implementation. The proposed method has been evaluated on three well-known datasets in this field, namely KTH, UCF, and HMDB51, and has achieved accuracies of 95.02, 93.14, and 74.28, respectively. It has also been compared with other popular methods under different conditions.

Keywords

Journal of Machine Vision and Image Processing

Articles in Press, Accepted Manuscript
Available Online from 28 March 2024

Files

Share

How to cite

Statistics