A computer vision-based mobile tool for assessing human posture: A validation study

Area of Science:

• Biomedical engineering
• Computer vision
• Human posture analysis

Background:

• Non-invasive postural assessment is crucial for monitoring deviations.
• Existing computer-based methods often require manual landmark identification.
• Mobile applications offer potential for accessible postural assessment.

Purpose of the Study:

• To present and validate NLMeasurer, a mobile application for automated postural assessment.
• To leverage computer vision and machine learning (PoseNet) for anatomical point identification.
• To calculate postural measures from identified anatomical points.

Main Methods:

• Twenty participants were photographed with and without surface markers.
• Postural measurements were computed using NLMeasurer and validated biophotogrammetry software.
• Agreement was assessed using t-tests and Bland-Altman analysis; reliability via ICC.

Main Results:

• NLMeasurer showed agreement with established biophotogrammetry software for postural measurements.
• Good inter- and intra-rater reliability was observed, particularly with surface markers.
• The application effectively calculates postural measures from identified points.

Conclusions:

• NLMeasurer is a valid tool for frontal view postural measurement.
• Surface markers enhance the accuracy and reliability of landmark identification.
• The app facilitates automated and accessible postural analysis.