Predicting human body dimensions from single images: A first step in automatic malnutrition detection

Malnutrition in children accounts for 45% of child deaths globally. Automatic mal-nutrition detection from digital photos serves as a decision support tool for early detection of malnutrition in rural areas. We study the feasibility of estimating body-shape characteristics from images of human body shapes as a first step in automatic malnutrition detection. We generate multi-view images of male and female bodies from rendered digital 3D scans of human bodies.
Using convolutional neural networks (CNNs), we estimated waist circumference and body height with a mean absolute error of 59 mm and 9 mm, respectively. The estimation error of waist circumference depends on viewpoint. We conclude that automatic malnutrition detection from single images seems feasible, provided one or more suitable viewpoints are used