Advancements in Multi-View Human Activity Recognition for Ambient Assisted Living

Abstract

This Ambient Assisted Living (AAL) uses technology to improve the well-being, autonomy, and security of seniors and disabled individuals. AAL services depend on Human Activity Recognition to detect human behaviors from sensor data. Compared to single-view models, multi-view HAR, which aggregates data from multiple sensors, provides improved accuracy and reliable detection. In this review, we assess recent progress in multi-view HAR for AAL, focusing on the value of lightweight deep learning models. We cover the development of HAR systems, as well as the emergence of multi-view datasets and highly efficient deep learning models that are explicitly designed for AAL scenarios. Furthermore, we address problems such as data synchronization, data protection, and adaptability, as well as possible remedies like sensor fusion and transfer learning. As a result, we also demonstrate the potential of multi-view HAR to increase AAL services and outlining future research trends in providing adaptive, efficient, and privacy-aware systems.