Human Activity Recognition Using Advanced Deep Learning Approaches

Human Activity Recognition (HAR) using image data has become an important research area, especially
in applications such as smart surveillance, fitness tracking, and behaviour analysis. This paper proposes an image-based
deep learning framework that combines Self-Supervised Learning (SSL) for feature extraction, Transformer-based
sequence modelling, convolutional neural networks (CNNs) for training, transformer-based architectures and
Generative Adversarial Networks (GANs) for data augmentation.Self-Supervised Learning (SSL) with Convolutional
Neural Networks (CNNs) to effectively classify human activities from static images. The SSL module enhances feature
learning by training the model to reconstruct masked regions of input images. The learned features are then used to
classify activities using a CNN classifier. The proposed approach achieves high accuracy on a labelled image dataset of
human activities. Evaluation results and visual analysis demonstrate the model’s strong generalization capabilities in
classifying diverse human actions.