Karlsruhe Institute of Technology
Automated Visual Inspection for Industrial Quality Assurance
In the manufacturing process, an inspection is aimed at determining if a product deviates from a set of given specifications. Depending on the demands and the product features, different visual inspection techniques are typically applied in industry, including but not limited to, laser triangulation, fringe projection, and deflectometry. The inspection setups have often several degrees of freedom including the position and orientation of the cameras and the illuminations, as well as their optical configurations. A manual setup design for inspecting complicated products, such as an engine block, requires a tedious trial-and-error process, associated with high costs and often non-optimal results. This talk will give an overview of different visual techniques used for industrial 3D scanning, starting from the traditional stereo reconstruction to specific technical solutions. Moreover, I will introduce the “Inspection Planning” problem for optimizing the sequence of sensor acquisitions to best scan a product. To quantify an inspection quality, we formulate the amount of information an acquisition delivers, as a reduction of our uncertainty about the product surface, using a Bayesian framework. This is then used an optimization cost fuction for the planning problem.This talk will also cover topics regarding the physically-based computer graphics simulation of images, and methods for quantification of the measurement uncertainty in an optical measurement.
Mahsa Mohammadikaji is currently a 3rd-year PhD student at the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany. Her PhD work is concentrated on applied computer vision for industrial inspection planning and product quality assurance. She received her master degree in artificial intelligence in 2014, and her bachelor degree in software engineering in 2012, both from Sharif University of Technology.
Watch video from here.