• 3D body scanning
• biometrics
• computer vision
• human body segmentation
• RGBD camera

Anthropometry plays an important role in a variety of fields such as product design, robotics, ergonomics, and medicine. For the latter, cardiac insufficiency is a widely spread disease, for which the human body limbs volume is a strong predictor. Their estimation is currently done by either tape measurements or body immersion. However, computer vision offers non-invasive, safe, cost-effective, and rapid methods capable of accurate body parameters evaluation.
In this work, a 3D body scanning apparatus is devised to estimate single body segment volume, thus overcoming the complexity of traditional evaluation while enhancing the accuracy of 3D body scanning. This system features a minimal setup consisting of two RGBD LiDAR cameras, acquiring a front and back depth image of a patient assuming a predetermined pose. A processing pipeline isolates and segments the body shape by the means of two DNNs, one for body landmark detection, the other for body segmentation. A common reference system provided by an ArUco marker before capture allows the unification of the two acquisitions into a single point cloud. The body shape is then reconstructed by fitting a template mesh to the unified point cloud with an ARAP deformation featuring a novel regularization term. The method has been experimentally assessed. A set of standardized anthropometric measurements is gathered on ten subjects and then compared to the measures extracted from the mesh after the scanning procedure.