Low-cost 3D human body reconstruction under limited views and girth measurement for the apparel customisation.

Abstract

To enhance the convenience of human body 3D modelling, this study proposes a low-cost method for 3D body reconstruction under limited views, aiming to easily acquire client body size information through smart phone photography. The human body photos of the front, side and back view are captured, and background removal is performed using the U²-Net human segmentation model. The PIFuHD model is utilised to obtain single-view point cloud patches, which are then mapped onto 2D images. A point cloud registration approach that incorporates regional segmentation and contour supervision is employed to reconstruct a complete human body model. B-spline curves are employed to fit key girths for obtaining perimeter measurements, and the effectiveness of girth measurements is validated using body data from 80 subjects. The results indicate that the proposed method exhibits closer proximity to manual measurements compared to 3D scanning, with average absolute errors within 2 cm.