Human Activity Recognition: From Sensors to Applications

Abstract

Human activity recognition (HAR) being a dynamic research topic in recent decades due to its high demand in countless applications, for instance, in healthcare, gaming, security and surveillance, and sports. Despite the amount of work contributed by the researcher to this well-researched field, there are still many challenging aspects and open issues that should be addressed in future works. In this paper, the current state-of-the-art in HAR from three holistic aspects is surveyed: sensors, models, and open challenges. First, we summarize the existing sensory systems, including sensor-based, vision-based sensors, and multimodal solutions. Next, the recent advances in HAR algorithms – from hierarchical fusion methods to handcrafted features to deep features, traditional machine learning algorithms to deep learning techniques – are discussed. Finally, the principal issues and challenges that should be addressed in future research are discussed.