Automated Cranial Mesh Reconstruction and Base Plane Definition via Landmark Detection and 2D Gaussian Splatting

다국어 초록 (Multilingual Abstract)

This study presents a low-cost, non-invasive pipeline for three-dimensional cranial measurement based on 360° video acquisition with a smartphone camera and 2D Gaussian Splatting. The purpose of this research is to enable reliable cranial reconstruction without reliance on expensive commercial imaging systems. In the proposed design, video recordings were captured under controlled conditions, and Structure-from-Motion techniques implemented in COLMAP were applied to estimate camera poses and reconstruct sparse point clouds. These outputs served as inputs to 2D Gaussian Splatting, producing a detailed three-dimensional mesh. A bounding-box procedure derived from the estimated camera poses was then used to remove background elements and retain only the cranial and upper-body region. To standardize measurements, two eyelid line landmarks were automatically detected on a designated reference frame, and their intersections with the mesh were computed to define a geometric base plane. Cutting along this plane isolated the cranial region above it, producing meshes ready for further morphometric analyses. Experimental results confirmed close alignment between recorded frames and reconstructed meshes, correct localization of landmark-based intersections, and stable mesh segmentation. The findings demonstrate that the proposed pipeline can generate analyzable cranial data suitable for assessing intracranial volume, cranial circumference, and asymmetry, offering a practical alternative to traditional clinical imaging systems.