EAAR: Efficient and Accurate Action Recognition model with enhanced spatio-temporal perception.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Video understanding is crucial for applications in security, transportation, and entertainment.
• Action recognition is a foundational task for video analysis, enabling tasks like gesture capture and label generation.
• Optimizing video data processing for efficiency and accuracy is key to improving user experience and operational efficiency.

Purpose of the Study:

• To propose an Efficient and Accurate Action Recognition (EAAR) model.
• To enhance the simultaneous sampling of temporal and spatial information in video data.
• To combine global and local features for a more characteristic action representation.

Main Methods:

• Developed an Efficient and Accurate Action Recognition (EAAR) model using a 2D Convolutional Neural Network (CNN).
• Introduced the Spatial-Temporal Fusion Module (STFM) for efficient spatial and temporal sensing.
• Implemented the Spatial-Temporal Optimized Module (STOM) for optimized feature representation with linearized complexity.

Main Results:

• The EAAR model achieved 97.7% accuracy in action recognition.
• The model demonstrated 98% accuracy in violence detection.
• Experimental results confirm the effectiveness of the proposed method in utilizing spatio-temporal features.

Conclusions:

• The proposed EAAR model effectively processes spatio-temporal features for accurate video understanding.
• The EAAR model, with its STFM and STOM components, offers an efficient and accurate solution for action recognition.
• The method shows significant potential for real-world applications requiring reliable video analysis.