Flexible registration method for light-stripe sensors considering sensor misalignments

Abstract

In many application areas such as object reconstruction or quality assurance, it is required to completely or partly measure the shape of an object or at least the cross section of the required object region. For complex geometries, therefore, multiple views are needed to bypass undercuts respectively occlusions. Hence, a multi-sensor measuring system for complex geometries has to consist of multiple light-stripe sensors that are surrounding the measuring object in order to complete the measurements in a prescribed time. The number of sensors depends on the object geometry and dimensions. In order to create a uniform 3D data set from the data of individual sensors, a registration of each individual data set into a common global coordinate system has to be performed. Stateof- the-art registration methods for light-stripe sensors use only data from object intersection with the respective laser plane of each sensor. At the same time the assumption is met that all laser planes are coplanar and that there are corresponding points in two data sets. However, this assumption does not represent the real case, because it is nearly impossible to align multiple laser planes in the same plane. For this reason, sensor misalignments are neglected by this assumption. In this work a new registration method for light-stripe sensors is presented that considers sensor misalignments as well as intended sensor displacements and tiltings. The developed method combines 3D pose estimation and triangulated data to properly register the real sensor pose in 3D space. © 2017 SPIE.