Acquisition and 3D registration of image sequences for structured light based free-form surface reconstruction

Accurate measurements on manufactured objects including free-form surfaces are of prime interest for (industrial) computer vision. However, resolution of standard CC'D sensors is a strong limitation for objects of large extent. To overcome this shortage, image sequences covering the areas of interest are acquired and registered. Making use of structured light for obtaining 3D information requires registration to rely on the extracted 3D data. Novelty of our highly accurate registration scheme for pairs of consecutive images lies in interpolation of the imaged surfaces before matching 3D points in the automatically determined overlap regions. allowing to significantly reduce the size of the search space for matching without spoiling registration accuracy. Further, the algorithm computes actual displacements between images, enabling to express the measured points in a reference frame. Application to spatial image sequences assess efficiency, robustness and accuracy (15 to 50μm. depending on object nature and size of image sequences).