A Novel Graph Neural Network based Approach for Human Activity Recognition

Human activity recognition (HAR) is an important research area that involves detecting and classifying human activities using various sensors. Recently, radar-based HAR systems have attracted the attention of researchers due to their superiority over other techniques. However, one of the main challenges in HAR is capturing the spatial dependencies among the micro-Doppler features, as the spatial arrangement of the activities and their features are necessary for accurate recognition. This paper introduces a novel graph neural network model for classifying human activities using superpixel gray-scale images constructed from the range and velocity profile obtained from human activity data. Our approach utilizes a k-NN (k-nearest neighbor) graph structure to represent the superpixels as nodes, enabling the model to learn spatial dependencies among the image features. We evaluate the performance of our model with a publicly available dataset using different GNN (Graph Neural Network) techniques, including GCN (Graph Convolution Neural Network), GAT (Graph Attention Network), and SGC (Simplifying Graph Convolution Network), and conduct a comparative study to determine the most effective approach for this task. The results indicate that GAT outperforms the other state-of-the-art techniques in accurately classifying human activities with competitive accuracy.