How precise is "Flying Triangulation"?

Last year we presented a novel optical measurement principle called Flying Triangulation: A single-shot sensor acquires a series of sparse 3D views which then are automatically aligned to each other in order to obtain a dense 3D model of the object. This enables a hand-guided, motion-robust measurement of object surfaces. We demonstrated the principal functionality and presented measurement examples. Now we discuss the accuracy of the sensor. Flying Triangulation incorporates three sources of errors: the physical measuring uncertainty, calibration errors of the single shot measurement, and - especially crucial for our sensor - registration errors. We first explain how to minimize the measurement uncertainty within each single 3D view, considering the physical limits. Then, we present efficient and robust algorithms to automatically align an acquired series of hundreds of sparse 3D views and discuss the resulting error. The local registration error is clearly below the achieved measurement uncertainty. Outlier detection and correction methods reduce the error propagation over all 3D views. Results based on simulated as well as on real data are displayed.