A Human Pose Similarity Evaluation using Joint Angles

Key Points

• The aim is to improve pose evaluation accuracy in dynamic environments using joint angles and cosine similarity.
• Developed a pose vector from MediaPipe joint coordinates.
• Applied cosine similarity for evaluating pose similarity.
• Conducted comparisons between angle-based and pixel-based assessments.
• Angle-based evaluations showed higher discrimination than pixel-based methods.
• Achieved the highest frames per second (FPS) under limited computational resources.
• Demonstrated the effectiveness of the proposed method in real-time dynamic posture analysis.