3D reconstruction, classification and mechanical characterization of microstructures

Official URL: http://192.168.1.20/record=b1226349 (Table of Contents)

Modeling and classifying 3D microstructures are important steps in precise micro-manipulation. This thesis explores some of the visual reconstruction and classification algorithms for 3D microstructures used in micromanipulation. Mechanical characterization of microstructures has also been considered. In particular, visual reconstruction algorithm (shape from focus - SFF) uses 2D image sequence of a microscopic object captured at different focusing levels to create a 3D range image. Then, the visual classification algorithm takes the range image as an input and applies a curvature-based segmentation method, HK segmentation, which is based on differential geometry. The object is segmented into surface patches according to the curvature of its surface. It is shown that the visual reconstruction algorithm works successfully for synthetic and real image data. The range images are used to classify the surfaces of the micro objects according to their curvatures in the HK segmentation algorithm. Also, a mechanical property characterization technique for cell and embryo is presented. A zebrafish embryo chorion is mechanically characterized using cell boundary deformation. Elastic modulus and developmental stage of the embryo are obtained successfully using visual information. In addition to these, calibrated image based visual servoing algorithm is experimentally evaluated for various tasks in micro domain. Experimental results on optical system calibration and image-based visual servoing in micropositioning and trajectory following tasks are presented.