Improved curvature-based registration methods for high-precision dimensional metrology

Abstract

Multiple measurements using various data acquisition systems are generally required to substancially enhance measurement accuracy, reliability and holisticity of freeform shapes. The obtained multiple measurement data of the shape are transformed and fused into a common coordinate system within a registration technique involving coarse and fine alignments. Standardized methods have been established for fine registration such as Iterative Closest Points (ICP) and its variants. For coarse registration, no conventional method has been adopted yet despite a significant number of techniques which have been developed in the literature to supply an automatic rough matching between data sets.The work presented in this paper proposes an improvement of registration techniques by the consideration of new discrete curvature parameters. Two main issues are addressed in this paper: the coarse registration and the fine registration. For coarse registration, two novel automated methods based on the exploitation of discrete curvatures are presented: an enhanced Hough Transformation (HT) and an improved Ransac Transformation. The use of curvature features in both methods aims to reduce computational cost. For fine registration, a new variant of ICP method is proposed in order to reduce registration error using curvature parameters. A specific distance considering the curvature similarity is combined with Euclidean distance to define the distance criterion used for correspondences searching. Additionally, the objective function is improved by combining the point-to-point (P-P) minimization and the point-to-plane (P-Pl) minimization with automatic weights. The algorithms are applied on simulated and real data performed by a computed tomography (CT) system. The obtained results reveal the benefit of the proposed improved curvature-based registration methods.