Automatic registration of 3D dental mesh based on photographs of patient's mouth

摘要

A time-consuming key point of virtual treatment planning in orthodontics is 3D model alignment. This paper proposes a fast and automatic registration process to set 3D dental models in occlusion using only a few photographs of patient's mouth. A set of reliable 3D-to-2D singular points correspondences is defined mixing colour and curvature analysis. Then, the rigid transform to bring the two jaws closer together estimated by minimising the reprojection error in multiple photographs. Refinement of point locations is incorporated into the optimisation process by introducing original strategies that limit the point displacement into targeted curves or areas to maintain an overdetermined system. Our approach is validated on several sets of real patients' data with a deviation of less than 1 mm and a maximum of 2° from the expert's reference.