A Structured and Methodological Review on Multi-View Human Activity Recognition for Ambient Assisted Living.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Ambient Assisted Living (AAL) systems aim to support elderly and disabled individuals using technology.
• Efficient Human Activity Recognition (HAR) is crucial for effective AAL, yet systematic comparisons of different approaches are lacking.

Purpose of the Study:

• To systematically review and compare single-view (SV) and multi-view (MV) Human Activity Recognition (HAR) approaches in the context of Ambient Assisted Living (AAL).
• To analyze the evolution of HAR systems from SV to MV architectures, focusing on deep learning models for AAL.

Main Methods:

• Comprehensive literature review analyzing benchmark datasets, feature extraction, and classification techniques for HAR.
• Examination of various machine learning and deep learning models, including CNNs, RNNs, LSTMs, TCNs, and GCNs.
• Discussion of lightweight transfer learning methods and sensor fusion strategies for resource-constrained AAL environments.

Main Results:

• Multi-view HAR architectures, particularly those employing advanced deep learning models, demonstrate enhanced accuracy and robustness compared to single-view systems in AAL.
• The study covers a wide array of models and techniques, highlighting their suitability for different AAL scenarios.
• Key challenges like data remediation, privacy, and generalization are identified, with potential solutions proposed.

Conclusions:

• Multi-view HAR systems represent a significant advancement for AAL, offering improved performance and adaptability.
• Future development should focus on intelligent, efficient, and privacy-preserving HAR solutions for AAL.
• The review provides a roadmap for researchers and developers in the field of AAL and HAR.