Presentation
A Comprehensive Methodological Framework for Anthropometric Head Shape Modeling Using Small Dataset
SessionLBR1: Late-Breaking Results
DescriptionDetailed anthropometric characterization of the human head can ensure optimal fit, comfort, and effectiveness of head-mounted devices. However, for laboratory-based, small, occupation-specific datasets, there is a lack of a reliable and systematic approach for head shape classification and modeling. Therefore, in this study, we proposed a six-step framework—pre-processing, feature extraction, feature selection, clustering, shape modeling, and validation—for head shape analysis using a small sample of 36 firefighter 3D head scans. We employed k-means and k-medoids clustering methods using both principal component analysis (PCA) inputs and the original set of measures. Additionally, five different curve fitting methods—NURBS applied to the average (NURBS AVG), NURBS least squares (NURBS LS), cubic spline applied to the average (Cubic Spline AVG), cubic spline by averaging the coefficients (Cubic Spline CAVG), and cubic spline least squares (Cubic Spline LS)—were employed to design the representative head shape of each cluster. The clustering results indicated that PCA-based k-means clustering outperformed k-medoids for both 1D- and shape-based inputs, highlighting k-means' superior performance with small sample sizes. Compared to the 1D-based input, the shape-based generated more clusters due to the greater amount of information provided by the 3D points. Furthermore, among the shape modeling methods, Cubic Spline LS displayed the lowest MSE (0.70 cm²) and computational time (0.14 s), whereas NURBS LS displayed the highest MSE (7.19 cm²) and substantially higher computational time (1.87 s). The proposed framework can aid in choosing the most adequate clustering and shape modeling techniques, especially when dealing with small datasets.
Event Type
Late Breaking Results
TimeTuesday, September 10th3pm - 3:10pm MST
LocationFLW Salon I