Intelligent Video Analytics for Human Action Recognition: The State of Knowledge

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human action recognition is crucial for intelligent video analytics.
• Existing methods include pose-based, tracking-based, spatio-temporal, and deep learning approaches.
• Visual transformers represent a recent advancement in this field.

Purpose of the Study:

• To provide a comprehensive overview of intelligent video analytics and human action recognition.
• To discuss the current state-of-the-art techniques and their limitations.
• To explore the potential of edge AI for real-time human action recognition.

Main Methods:

• Review of pose-based, tracking-based, and spatio-temporal methods.
• Analysis of deep learning-based approaches, including convolutional neural networks and recurrent neural networks.
• Inclusion of visual transformers as a novel deep learning technique.
• Discussion of edge AI architectures for efficient computation.

Main Results:

• Detailed comparison of various human action recognition techniques.
• Identification of challenges and limitations in current methods.
• Highlighting the feasibility of edge AI for real-time applications.

Conclusions:

• Human action recognition is a rapidly evolving field with diverse methodologies.
• Deep learning and visual transformers offer promising avenues for improved accuracy.
• Edge AI architectures are key to enabling real-time, on-device human action recognition.