Methods to improve the accuracy of registration for multimodal inspection of mechanical parts

Registration refers to the process of aligning corresponding features in images or point data sets in the same coordinate system. Multimodal inspection is a growing trend wherein an accurate measurement of the part is made by fusing data from different modalities. Registration is a key task in multimodal data fusion. The main problem with high-accuracy registration comes from noise inherent in the measurement data and the lack of the one-to-one correspondence in the data from different modalities. We present methods to deal with outliers and noise in the measurement data to improve registration accuracy. The proposed algorithms operate on point sets. Our method distinguishes between noise and accurate measurements using a new metric based on the intrinsic geometric characteristics of the point set, including distance, surface normal and curvature. Our method is unique in that it does not require a-priori knowledge of the noise in the measurement data, therefore fully automatic registration is enabled. The proposed methods can be incorporated into any point-based registration method. It was tested with the traditional ICP (Iterative Closest Point) algorithm with application to the data registration among point, image, and mesh data. The proposed method can be applied to both rigid and non-rigid registration.