mModPoEs: Multimodal posture estimation and feedback-driven correction of load-bearing human movements using wearable...

Area of Science:

• Biomedical Engineering
• Computer Science
• Rehabilitation Science

Background:

• Improper spinal posture during daily activities is a significant cause of musculoskeletal disorders, including chronic back pain and disc degeneration.
• Current posture assessment methods may lack comprehensive data integration for effective spinal health support.

Purpose of the Study:

• To present a multimodal posture estimation and feedback framework integrating wearable sensor data and computer vision.
• To enhance the accuracy of posture classification for supporting spinal health and preventing musculoskeletal issues.

Main Methods:

• Integrated data from Inertial Measurement Units (IMUs) and flex sensors for postural angle quantification.
• Utilized multi-view video analysis with the MediaPipe framework for visual feature extraction.
• Employed machine learning algorithms including logistic regression, decision tree, random forest, KNN, and SVM on data from 40 healthy young adults.

Main Results:

• The Random Forest algorithm demonstrated effective performance across various activities (sitting, standing, walking) for both genders.
• Achieved accuracy rates of 75% for sitting, 95% for standing, and 63% for walking postures.
• The multimodal approach, combining wearable and visual data, significantly enhanced posture classification accuracy.

Conclusions:

• Integrating wearable sensors and computer vision modalities improves posture classification accuracy.
• The developed framework provides a methodological foundation for future multimodal, feedback-based posture assessment systems.
• Preliminary findings suggest potential for reducing musculoskeletal disorders through enhanced posture monitoring.