Single camera multi-view anthropometric measurement of human height and mid-upper arm circumference using linear...

Area of Science:

• Anthropometry
• Medical Imaging
• Public Health

Background:

• Manual anthropometry is crucial for malnutrition assessment but requires training, limiting its use in resource-constrained areas.
• Photogrammetry, utilizing affordable digital cameras, is emerging as a viable alternative for anthropometric data collection.
• Accurate anthropometric measurements are vital for monitoring nutritional status and guiding public health interventions.

Purpose of the Study:

• To assess the accuracy of estimating height and mid-upper arm circumference (MUAC) using linear regression on photogrammetric distance measurements.
• To determine optimal combinations of photographic views for accurate anthropometric estimations.
• To evaluate simplified photogrammetric approaches for reduced complexity and expertise requirements.

Main Methods:

• Thirty-one adults were photographed from five views, with distances calculated via camera calibration and reference objects.
• Linear regression was applied to photogrammetric distances to estimate height (head-to-floor and bounding box) and MUAC (arm width).
• Estimates were compared across different view combinations and against existing photogrammetric techniques.

Main Results:

• Linear regression estimates showed smaller mean absolute differences (MAD) compared to other photogrammetric methods.
• Optimal view combinations (front and side) yielded reliable estimates, with MAD for height around 10-11 mm and for MUAC around 7 mm.
• Camera calibration further improved accuracy, reducing MAD by 2-3 mm.

Conclusions:

• Photogrammetry with linear regression accurately estimates height and MUAC, meeting nutritional assessment requirements.
• Future research will focus on automating landmark detection and validating the method on diverse, including undernourished, populations.
• This approach holds potential as a practical tool for nutrition assessment by non-expert users.