Alignment of dental depth images from an intraoral scanner

Abstract

This paper proposes an automated alignment procedure of dental depth images captured by intraoral scanners like the Microsoft Kinect. The proposed procedure identifies an acquired order from initial scanning data and attempts to compute the registration sequence for aligning all dental depth images into a completed model. A core part of the proposed procedure is an algorithm that computes the similarity of each dental depth images, considering only the adjacent dental depth images. The algorithm was designed carefully by considering two major technological requirements of the problem: accuracy and efficiency. To satisfy the accuracy requirement, the proposed algorithm uses the concept of feature vector based on fast point feature histogram (FPFH) to compute the similarity index of each dental depth images. In this phase, the computation of nonadjacent dental depth image was excluded by the axis aligned bounding box (AABB) detection, satisfying the required efficiency. The data used in experiment is consisted of 1,021 dental depth images of mandibular teeth.