Key points are not available for this paper at this time.
ABSTRACT Detailed anthropometric characterization of complex shapes of human heads can ensure optimal fit, comfort, and effectiveness of head-mounted devices. However, there is a lack of a reliable and systematic approach for head shape classification and modeling for laboratory-based, small, occupation-specific datasets. Therefore, in this study, we proposed a streamlined framework comprising six steps—pre-processing, feature extraction, feature selection, clustering, shape modeling, and validation—for head shape classification and modeling. We collected 36 firefighter 3D head scans and implemented the framework. Different clustering techniques, such as k-means and k-medoids, were evaluated using the squared Euclidean distance of individual head shapes from their cluster centroid. Furthermore, five variations of NURBS and cubic spline methods were assessed to design the representative head shape of each cluster, and their accuracy was evaluated using mean square error (MSE) values. The clustering results indicated that k-means provide better metrics than k-medoids. Among the shape modeling methods, cubic spline least squares displayed the lowest MSE (0.70 cm 2 )and computational time (0.14 s), whereas NURBS least squares displayed the highest MSE (7.19 cm 2 ). Overall, the framework with k-means clustering and cubic spline least squares modeling techniques proved to be the most efficient for small datasets.
Wei et al. (Mon,) studied this question.