Image Matching and Surface Registration for 3D Reconstruction of a Scoliotic Torso

Abstract

The focus of this research is a hierarchical image matching strategy and a multiple surface registration technique for 3D reconstruction of a scoliotic torso. Scoliosis is a deformity of the human spine most commonly occurring in children. After being detected, periodic examinations via x-rays are traditionally used to measure its progression. However, due to the increased risk of cancer, non-invasive and radiation-free scoliosis detection and progression monitoring methodologies are being researched. For example, quantifying the scoliotic deformity through the torso surface is a valid alternative because of its high correlation with the internal spine curvature. This work proposes a low-cost, multi-camera photogrammetric system for semi-automated 3D reconstruction of a torso surface with a sub-millimetre level accuracy. The paper first describes the system design and calibration for optimal accuracy. It then covers the reconstruction and registration procedures giving insights into the hierarchical image matching strategy and the multiple surface registration technique. Final accuracy is evaluated through the goodness of fit between the reconstructed surface and a more accurate set of points measured by a coordinate measuring machine.