Computational Illumination in Microscopic 3D Metrology. Martin Lenz
Sprache des Titels:
Within the last two decades, camera-based microscopic 3D metrology became a well-established method in contactless object inspection tasks for industry. In particular, a computer vision approach called shape from focus emerged when surface color and roughness of micro-structures are of interest in a 3D measurement process. In literature, a variety of shape from focus methods for reflected-light microscopy already exist. However, all methods deliver inaccurate or sparse reconstructions for mainly two reasons: First, untextured and homogeneous image regions do not respond correctly regarding applied focus measures. Second, due to the classic acquisition process using a co-axial light source with respect to the observing camera, over-saturated image regions frequently appear because of specular reflections. On the other hand, averted scene geometries only reflect a fraction of light - resulting in under-exposure.
In this talk, I will present new methods from the field of computational photography that automatically optimize the image acquisition process in microscopic 3D metrology and hence tackle the mentioned problems by increasing the image quality of input data. Two different approaches are proposed: One using a ring-light with individually adjustable light segments. The second approach incorporates a projector as co-axial light source to the camera. This allows to (a) actively project a texture onto homogeneous scene regions, (b) locally adjust illumination intensities to avoid saturation effects in an image and furthermore (c) to radically reduce the number of required image acquisitions for a 3D reconstruction.