Multi-view 3D Reconstruction Using Monocular Microscopic Images

Microscopic 3D reconstruction, as a core technique of micro-vision system, has a very important application value in precision manufacturing. However, because of obvious microscopic image degradation, 3D reconstruction from multi-view images needs to be improved from many aspects when it is applied in micro field.The goal of this paper is to establish a simple and efficient microscopic 3D reconstruction system with monocular microscopy and apply it to the surface reconstruction of micro-scale objects, thus providing essential references for microscopic inspection. The main tasks and achievements are as follows:1. Comparative study of image focusing evaluation function. Image focusing evaluation function is one of the important basis to make micro-vision system focus rapidly. Nine specific image focusing evaluation functions of three kinds have been studied. How the robustness of different kinds of evaluation function change when manification and noise intensity change was analyzed through experiments. The focusing quality evaluation scheme of microscopic image collection was finally determined with the study results.2. The study of the keypoint detection and matching algorithm of images. Keypoint detection and matching are crucial elements of reconstruction precision. This chapter introduced constraint based on local geometric consistency of transform on the base of performance comparison of four usual keypoint detection algorithms and then created a new outlier removal algorithm. This algorithm can obtain more matching points and meanwhile reduce the matching error substantially. Also, it can eliminate matching mistakes near by epipolar line which cannot be inspected by RANSAC and the like.3. Develop a set of rounded microscopic 3D reconstruction system. The microscopic 3D reconstruction system hardware in this paper consists of 4-DOF motion platform and microscopic imaging system, so it can collect multi-view microscopic images easily. The nucleus module of the system software integrated microscopic image capturing method, microscopic image keypoint detection and matching algorithm, and further combined Structure From Motion frame to achieve efficient microscopic 3D reconstruction.