A comprehensive methodological framework for 3D head anthropometric shape modeling of a small dataset

Efficient data analytics methods are essential to characterise occupation-specific anthropometric head shapes for developing well-fitted head-mounted devices. However, classifying and modelling 3D head shapes for small population groups remains challenging due to limited data and systematic approaches. This study proposes a streamlined six-step framework using 3D head scans from 36 firefighters (18 males, 18 females). We evaluated K-means and K-medoids clustering and four shape modelling methods-NURBS, NURBS least squares (LS), Cubic Spline, and Cubic Spline LS-and validated the predicted head shape against NIOSH, ANSUR II, CAESAR, and US Army databases. Results showed K-means outperformed K-medoids (28% lower distances). Surface mapping-based clustering was 35% more accurate than PCA-based clustering. Cubic Spline LS achieved the lowest mean squared error (0.70 cm²) and fastest computation (0.14 s), performing better than NURBS LS (7.19 cm² and 1.87 s). Overall, surface mapping, K-means clustering, and Cubic Spline LS methods provided more accurate head shapes for our studypopulation groups.