Effectiveness of Vision Transformers in Human Activity Recognition from Videos

Abstract

Human Action Recognition (HAR) has got the attention of computer vision domain researchers due to its wide variety of applications like surveillance, behavior detection, sports action monitoring, and elderly monitoring. Due to the huge amount of data, the Deep Learning-based method is widely used in HAR compared to the Machine Learning-based approach. This study explored the various Deep Learning and pre-trained Deep Learning models in HAR. In the pre-trained model, we do not require to train the model from scratch, which is already trained on huge data. This study explored the recent pre-trained Deep Learning model to classify action accurately. This study helps the researcher to evaluate the benefit of the latest Vision Transformer model in the domain of HAR.UCF 50 action dataset is used in this study to examine the effectiveness of the Vision Transformer model in HAR. On UCF 50 action dataset, we have achieved 94.70% accuracy using the Vision Transformer model variant.