OctaCoG alignment: A fast and reliable alternative to iterative alignment schemes

For metrology tasks in the context of industrial quality inspection and non-destructive testing, tactile measurements are more and more superseded by imaging techniques, such as optical scanning and Computed Tomography. Data fusion of such very differently acquired data plays a key role in combining their strengths and mitigating their weaknesses. This requires a highly accurate alignment of the provided data sets. Classical alignment schemes turn out to be prone to error for this task due to differences in resolution and contributing surfaces from these fundamentally different imaging techniques. Our approach reduces such errors for the special case of aligning complete representations of both types of acquisition. With OctaCoG, we present a multi-modality registration method for complete object representations. It operates on point clouds of the exterior surface and does not require user interaction for tie point identification. A combination of principal component analysis and barycenter computation provides a characteristic shape per data set. This shape consists of eight feature points, and allows for an alignment outperforming the accuracy of classical approaches and, as a positive side effect, surpassing their run time.