A Review on Human Activity Recognition Using Vision-Based Method.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Human Activity Recognition (HAR) is crucial for applications like video surveillance, healthcare, and human-computer interaction (HCI).
• Vision-based HAR relies on analyzing subject actions and environmental context from visual data.
• Advances in activity representation and classification are key to improving HAR system performance.

Purpose of the Study:

• To review state-of-the-art approaches in vision-based human activity recognition.
• To provide a chronological overview of activity representation methods, from global to local and depth-based.
• To categorize and review prevalent classification methods, including template-based, discriminative, and generative models.

Main Methods:

• Systematic review of literature on activity representation and classification.
• Chronological sorting of representation techniques.
• Categorization of classification methods into template-based, discriminative, and generative models.
• Introduction of representative and available datasets for HAR research.

Main Results:

• Identified a research trajectory in activity representation: global -> local -> depth-based.
• Classified HAR methods into distinct categories for representation and classification.
• Presented a taxonomy of existing literature, linking methods to datasets used.
• Highlighted key datasets for benchmarking HAR algorithms.

Conclusions:

• The review offers a structured overview of vision-based HAR methods.
• It facilitates comparison of different approaches and datasets.
• Identified future research directions in activity representation and classification for enhanced HAR systems.