A parametric framework for population-specific 3D human body shape reconstruction using SMPL

To address the issue of inadequate adaptability of the parametric human template model in representing the body shapes of specific populations, this study proposes a 3D model shape optimisation method tailored for specific populations based on the SMPL model. Firstly, multiple sets of human models are generated based on SMPL, and body sizes are extracted to construct a generated model dataset; simultaneously, a dataset of body sizes for Chinese adult males is established. Then, the body shape differences between the generated models and the target population are calculated, and the kernel function of the GPR algorithm is optimised to establish an end-to-end mapping from the body sizes to the shape parameters of the target population. The results indicate that the optimised mapping model reduces the MAE of human body sizes by 27.7%, significantly enhancing the applicability of the SMPL model to other population-specific characteristics.