Guided anchor data extraction and registration from incomplete and noisy scanned surface

摘要

Accurate surface registration plays a vital role in computer vision. In order to increase the accuracy of the registration, this paper proposes a general computational framework to extract the anchor data from the scanned surface for registration. A target surface is provided to guide the extraction and the registration. First, the anchor data on the target surface is manually highlighted according to prior knowledge. After that, a fully automatic system will extract the anchor data from each scanned surface and align the scanned surface to the target surface based on the anchor data. The extraction and the registration are performed iteratively to locate the correct anchor data and increase the accuracy of the registration. In each iteration, extraction derives anchor data to enable the accurate registration, and the registration result is used to further refine the anchor data. In order to prevent infinite loops in the algorithm, orientation correction is proposed to perturb the transformation after the registration and before updating the anchor data. Experiments on different engineering parts demonstrate its robustness and convergency. Compared with traditional methods, our framework increases the registration accuracy when dealing with incomplete and noisy data.