Based On Binocular Stereo Vision Melon Seeds 3 D Special Form Of Detection Research

Computer vision for automatic selection of agricultural products has been applied into some areas to certain extent. It distinguishes objects of different types, barley and wheat for instance, according to their differences on shapes and veins and then sorts them efficiently. For objects of the same types, their quality can be examined and measured by two-dimensional colored images that provide information like their color distribution and sizes. However, for flat granules of irregular shapes whose size is small and distribution of color and luster is asymmetrical, like watermelon seeds, there are some limitations and weaknesses to sort them by current approaches based on their size, length-width ratio and color distribution without considerations on their warpage.This thesis explores on the extraction and representation of the flat granules’ warpage by using computer vision technology and puts forward a method to represent the degree of warpage of flat granules’surfaces. In doing so, it is hoped to eliminate the limitations of current approaches. The research contents are as follows:Ⅰ. On the basis of research framework of this thesis, a research is to be done on Binocular Stereo Vision. A comparison of experiments’ results and theoretical analysis enable the author to come up with a standard optimum proposal based on OpenCV which aims to improve the accuracy and efficiency of stereo-camera.Ⅱ. As the results of stereo matching exert direct influences on the computing of three-dimensional point coordinates, stereo pictures to be matched and gradation histogram are preprocessed by denoising to improve the accuracy of stereo matching. After analysis of stereo matching and checking by experiments, a matching method is to be proposed for flat granules like melon seeds.Ⅲ. The author of this thesis analyzes the point cloud data from three-dimensional reconstruction and repairs holes in melon seeds caused by not having matching points with a triangulation method. Analysis on the point cloud data that are produced after repairing holes, a method integrating Skeleton line and partial Curvature Scale Space is raised to measure the degree of warpage and shape of melon seeds. Finally, the author visualizes the point and cloud data from three-dimensional reconstruction three-dimensionally based on OpenGL in order to satisfy needs such as the tests of three-dimensional reconstruction results and spot checks in production.