UWB Radar-based Human Indoor Activity Recognition Using StarNet Network

Human Activity Recognition (HAR) using radar sensors offers significant advantages due to their ability to operate in varying lighting conditions and through obstacles, preserving privacy. This paper proposes a novel HAR system that utilizes a multi-radar setup in conjunction with the StarNet model integrating convolutional neural networks (CNNs) and transformer technologies. The multi-radar system enhances the generalizability and fidelity of indoor activity data. Meanwhile, StarNet enhances traditional CNNs by incorporating the star operation, which efficiently maps features into high-dimensional spaces, achieving superior performance with an accuracy of 95.8%, and outperforming other models by at least 5.0%. This system is particularly beneficial in healthcare, eldercare, and smart home environments, providing robust and reliable activity monitoring.