Computer Vision Measurements for Automated Microrobotic Paper Fiber Studies

The mechanical characterization of paper fibers and paper fiber bonds determines the key parameters affecting the mechanical properties of paper. Although bulk measurements from test sheets can give average values, they do not yield any real fiber-level data. The current, state-of-the-art methods for fiberlevel measurements are slow and laborious, requiring delicate manual handling of microscopic samples. There are commercial microrobotic actuators that allow automated or tele-operated manipulation of microscopic objects such as fibers, but it is challenging to acquire the data needed to guide such demanding manipulation. This thesis presents a solution to the illumination problem and computer vision algorithms for obtaining the required data. The solutions are designed for a microrobotic platform that comprises actuators for manipulating the fibers and one or two microscope cameras for visual feedback.The algorithms have been developed both for wet fibers, which can be treated as 2D objects, and for dry fibers and fiber bonds, which are treated as 3D objects. The major innovations in the algorithms are the rules for the micromanipulation of the curly fiber strands and the automated 3D measurements of microscale objects with random geometries. The solutions are validated by imaging and manipulation experiments with wet and dry paper fibers and dry paper fiber bonds. In the imaging experiments, the results are compared with the reference data obtained either from an experienced human or another imaging device. The results show that these solutions provide morphological data about the fibers which is accurate and precise enough to enable automated fiber manipulation. Although this thesis is focused on the manipulation of paper fibers and paper fiber bonds, both the illumination solution and the computer vision algorithms are applicable to other types of fibrous materials