Human Action Recognition using an Ensemble Deep Learning Model for Video Datasets

Abstract

Human Action Recognition is used to analyse the videos to identify the actions performed by humans. In recent years, it has gained much popularity due to its large domain of applications presented in various fields. Several research contributions are available in this area but still the requirement is to achieve good results for various challenging datasets and limited hardware resources.  In order to overcome these issues, an ensemble deep learning model is proposed in this paper based on custom convolutional neural network (CNN). ResNet50 is merged with a handcrafted CNN, to identify human actions in challenging video datasets. This model is trained and tested on UCF-101 and HMDB-51 datasets and gained very good results. The experimental results showcased that the proposed model outperforms some recent works in this domain.