Robust pattern indexing methods for "Flying Triangulation"

The measurement principle "Flying Triangulation" (S. Ettl et al., Appl Opt 51 (2012) 281-289) enables a comfortable and motion-robust acquisition of 3D data. The handheld sensor, based on light sectioning, is freely moved around the object, while generating a series of sparse 3D views. These views are aligned to each other "on the fly" and the current measurement result is displayed in real time. A dense 3D model is generated within a few seconds. An essential part of this process is a correct identification of the projected lines. If an object part is observed within the measurement volume of the sensor, so-called "regions of uniqueness" guarantee that no errors occur. However, object parts outside the measurement volume lead to incorrect indexing. Since the sensor is freely hand-guided, such outliers are unavoidable. An automatic removal a posteriori is commonly impossible, since the outliers might intersect with correct data. We will present methods to overcome this problem, by combining hardware- and software-based approaches. Measurement examples will be given to demonstrate the improvement.