Human action recognition based on HOIRM feature fusion and AP clustering BOW

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Human action recognition is crucial for surveillance, human-computer interaction, and robotics.
• Existing methods often struggle with variations in camera viewpoints and distances in complex environments.
• The need for robust and accurate feature representation and classification models is paramount.

Purpose of the Study:

• To propose an improved human action recognition method.
• To enhance feature robustness and classification accuracy using feature fusion and an advanced clustering model.
• To validate the method's effectiveness on standard action recognition datasets.

Main Methods:

• Proposed a novel Histogram of Oriented Interest Region Motion (HOIRM) feature extraction method.
• Fused HOIRM features with 3D Histogram of Oriented Gradients (HOG) and 3D Histogram of Optical Flow (HOF) using a cumulative histogram.
• Developed a Bag-of-Words (BOW) model utilizing Affinity Propagation (AP) clustering for joint feature description and classification.

Main Results:

• Achieved an average recognition rate of 95.75% on the KTH dataset.
• Attained an average recognition rate of 88.25% on the UCF dataset.
• Demonstrated superior performance compared to methods using individual features (HOIRM or 3D HOG+3D HOF) and other existing approaches.

Conclusions:

• The proposed HOIRM feature fusion combined with the AP-based BOW model significantly improves human action recognition accuracy.
• The method exhibits enhanced robustness against variations in camera view angle and distance.
• The approach provides a more effective solution for complex human action recognition tasks.