Classification of Multiple Human Activities using Deep Learning

• Main Authors: Sangeetha, K., Vishnu Raja, P., SS, Lokeswaran, PG, Kishore Kumar, Kumaar S, Mohan
• Format: Conference Proceeding
• Language: English
• Subjects: Activity Recognition; Benchmark testing; Computational modeling; Convolutional Neural Network; Deep learning; Evaluation Metrics; Human Activity Classification; Human Activity Recognition; Recurrent neural networks; Reviews; Surveillance; Transfer learning
• ISSN: 2767-7788
• DOI: 10.1109/ICICT60155.2024.10544732

Human activity classification has emerged as a pivotal area of research within the field of deep learning, driven by its manifold applications spanning from healthcare to surveillance and human- computer interaction. This review paper provides an extensive examination of recent advancements in human activity classification utilizing deep learning techniques, shedding light on current challenges and promising future directions. We present a thorough analysis of the key components within the Convolutional neural networks (CNNs), recurrent neural networks (RNNs), and attention mechanisms are examples of deep learning paradigms that have significantly contributed to the accuracy and robustness of human activity classification models. The utilization of multi-modal data sources, including sensor data and image sequences, is explored, emphasizing the fusion of information from diverse sensors to enhance classification performance. Addressing the perennial challenge of data scarcity, we discuss the evolution of semi- supervised learning and transfer learning methods as strategies to maximize model generalization using limited labeled data. Furthermore, we investigate the impact of the ever-growing datasets and benchmark standards on the advancement of this field, enabling fair comparisons and benchmarking of models. This study also provides insights into ongoing research on real-time human activity recognition, where deep learning models are applied to interactive systems and robotics. The potential for enhancing interpretability and model explainability is explored, paving the way for more transparent and accountable AI systems in this context. Finally, this study highlights the significance of ethical considerations and privacy concerns in the development and deployment of human activity classification models and propose guidelines for responsible AI in this domain. This comprehensive review serves as a reference point for development and deployment of human activity classification models and propose guidelines for responsible AI in this domain.