SkelResNet: Transfer Learning Approach for Skeleton-Based Action Recognition

Skeleton-based action recognition is an increasingly popular research area in computer vision that analyzes the spatial configuration and temporal dynamics of human action. Learning distinctive spatial and temporal features for skeleton-based action recognition is one of the main challenges in this field. For this purpose, various deep learning methods such as CNN, RNN, GCN and Transformer have been used in the literature. Although these methods can achieve high performance, they require high computational costs and large datasets due to their complexity. Transfer learning is an approach that can be used to overcome this problem. In transfer learning, a pre-trained model can be fine-tuned for a new task. In this way, the computational cost can be reduced and high performance can be achieved with less data. In this study, SkelResNet architecture is designed based on the pre-trained ResNet101 model. Four different image representations were created using skeletal data to meet the input requirements of the SkelResNet architecture. Experimental studies have shown that SkelResNet outperforms CNN-based methods in the existing literature in action recognition.