Statistical body shape models allow for an intuitive generation of a subject-specific body shape with a few given individual’s body characteristics such as stature and body weight. This parametric behavior of the model also allows systematic analyses of effects of each parameter on the body shape.
Two statistical methods are generally used for developing the models: principal component analysis (PCA) and multivariate regression analysis. PCA reduces the high dimensionality of the original datasets so that a body information (3D coordinates of vertices, body landmarks, joint locations, and anthropometric data, etc) can be represented with a few value, a.k.a., “PC scores”. These PC scores, which efficiently accounts for the original data distribution, then are associated meaningful parameters like stature and BMI using the multivariate regression analysis.
Dr. Matt Reed and I have come up with various methods to make the models more realistic and applicable. The featured ideas are as follows:
- Non-rigid registration of scans while preserving anatomical homologies across the scans using RBF and Implicit-surface fitting methods
- Bootstrapping standardization process to improve the model quality using a rapid PC-based fitting method
- Advanced regression analysis to describe the expected nonlinearity in body shape change with different postures
These models will find much more applications including anthropometrics, ergonomics, vehicle design, product development, medical diagnosis, apparel design, and etc.