Human Pose Estimation for Real-World Deployment: A Review of Methods, Systems, and Applications

Area of Science:

• Computer Vision
• Machine Learning
• Robotics

Background:

• Human pose estimation (HPE) is crucial for human-centered visual understanding across diverse applications.
• Current benchmark performance often fails to translate to real-world deployment due to various practical constraints.

Purpose of the Study:

• To review Human Pose Estimation (HPE) from a deployment-oriented perspective, addressing the gap between benchmark performance and real-world reliability.
• To analyze the end-to-end HPE pipeline and identify key challenges and optimization strategies for practical systems.

Main Methods:

• Structured narrative synthesis of representative literature on HPE.
• Analysis of the end-to-end HPE pipeline, from data acquisition to task-level decision-making.
• Review of CNN-based, Transformer-based, temporal modeling, and lightweight optimization methods.

Main Results:

• Identified major benchmark-to-deployment gaps including data, system, task, and temporal issues.
• Analyzed the full HPE pipeline, highlighting factors influencing performance in real-world settings.
• Evaluated various HPE methods based on deployment cost, optimization potential, and application suitability.

Conclusions:

• HPE deployment requires considering practical sensing and system constraints beyond benchmark accuracy.
• Future research should focus on deployment-aware evaluation, system-level optimization, efficient temporal modeling, and application-level reliability for robust HPE.