Deep learning-based human body pose estimation in providing feedback for physical movement: A review

Area of Science:

• Computer Science
• Human-Computer Interaction
• Biomedical Engineering

Background:

• Human pose estimation is crucial for analyzing movement and providing skill-improvement feedback.
• Existing research utilizes various methods for pose estimation and movement assessment.

Purpose of the Study:

• To review the current research status and identify gaps in human body pose estimation and movement assessment.
• To analyze methods for pose estimation, movement assessment, user feedback, and evaluation.

Main Methods:

• Systematic literature search of Scopus and Web of Science databases.
• Analysis of 45 articles using bottom-up and top-down approaches.
• Categorization of methods for pose estimation, movement assessment, feedback, and evaluation.

Main Results:

• Convolutional Neural Networks (CNNs) are commonly used for pose estimation.
• Movement assessment methods include mathematical models, rule-based systems, and machine learning.
• Feedback is predominantly visual (verbal and nonverbal); evaluation uses datasets and human participants.

Conclusions:

• Pose estimation libraries are vital for field advancement.
• The effectiveness and selection criteria for movement assessment methods in new contexts are not well-defined.
• Further research is needed on feedback prioritization and the impact of erroneous feedback.