Vision Transformers-based Hand Gesture Classification

Hand gesture recognition (HGR) is a hot topic in machine learning and image processing communities. HGR is also vital for some Human-Computer Interaction (HCI) applications. Up to now, traditional machine learning approaches and deep convolutional neural networks (CNN) have been applied to HGR. Although these methods perform well enough on HGR, in this paper, we used a recent model namely vision transformer (ViT) on HGR. ViT is developed for improving the performance of CNN. ViT has a similar architecture to CNN but it has also different layers for the classification task. We used the ViT in a transfer learning fashion and applied it to the NTU hand gesture dataset. A holdout cross-validation test approach is considered in experiments and classification accuracy is used as a performance measure. The experimental works produce a 96.4% accuracy score and a comparative study with CNN models is carried out. Results show that the proposed model has potential in HGR.