Integrating OpenPose and SVM for Quantitative Postural Analysis in Young Adults: A Temporal-Spatial Approach

Area of Science:

• Computational Science
• Biomechanical Analysis
• Machine Learning Applications

Background:

• Traditional posture monitoring often relies on invasive sensors, limiting real-time applications.
• Significant potential exists to improve postural control quantification using readily available walking videos.

Purpose of the Study:

• To develop and validate a noninvasive computational method for accurate postural analysis.
• To integrate OpenPose deep learning with Support Vector Machine (SVM) for robust gait analysis.

Main Methods:

• Utilized OpenPose for deep learning-based extraction of Dynamic Joint Nodes Plots (DJNP) and iso-block postural identity images.
• Applied Temporal and Spatial Regression (TSR) models to extract key features for SVM classification.
• Analyzed walking data from 35 young adults in controlled experiments.

Main Results:

• Achieved an overall classification accuracy of 0.990 and a Kappa index of 0.985.
• Demonstrated effective differentiation of left and right skews using the ratio of top angles (TAR) and ratio of bottom angles (BAR) with high AUC values.
• Validated the efficacy of the integrated OpenPose-SVM approach for precise, real-time postural analysis.

Conclusions:

• The integration of OpenPose and SVM offers a highly accurate and robust noninvasive solution for postural control quantification.
• This method shows promise for revolutionizing clinical applications in rehabilitation and sports science.
• Future research will explore broader demographics and advanced machine learning models for enhanced adaptability.