SOCA-PRNet: Spatially Oriented Attention-Infused Structured-Feature-Enabled PoseResNet for 2D Human Pose Estimation

Area of Science:

• Computer Vision
• Machine Learning

Background:

• 2D Human Pose Estimation (HPE) is crucial for understanding human behavior in computer vision (CV).
• Deep learning has advanced 2D HPE, but challenges like occlusion and poor lighting persist.
• Optimizing 2D HPE is vital for accurate joint identification and classification.

Purpose of the Study:

• To introduce the Spatially Oriented Attention-Infused Structured-Feature-enabled PoseResNet (SOCA-PRNet) for enhanced 2D HPE.
• To improve accuracy in 2D HPE without a significant increase in model parameters.

Main Methods:

• Developed SOCA-PRNet, integrating Spatially Oriented Attention (SOCA) and Global Context Blocks (GCBs) with ResNet34.
• Employed deep learning for automatic feature learning and model generalization.

Main Results:

• Achieved a Percentage of Correct Keypoints at 0.5 (PCKh@0.5) of 90.877 at a 50% threshold.
• Attained a Mean Precision (Mean@0.1) score of 41.137.
• Outperformed existing state-of-the-art approaches in empirical evaluations.

Conclusions:

• SOCA-PRNet demonstrates superior performance in 2D HPE.
• The model's efficiency and accuracy show potential for real-world applications like robotics, gaming, and human-computer interaction.