Research On Recognizing And Separating Method For The Surface Flatness Of Irregular-granular Objects

In recent years, machine vision technology has been increasingly used in the field of agricultural product testing. At present, apple, mango and other fruits and vegetables can be successfully sorted by size, color, aspect ratio and other characteristics. However, due to slightly irregular surface, small size, little surface information and the irregular contour, it's difficult to characterize the irregular-granular objects'surface flatness which contains three-dimensional information.Aimed to resolve the difficult problem of obtaining and separating the surface flatness of irregular-granular Objects, this thesis focused on the surface flatness extraction of the irregular-granular objects, and designed an automatic method for identification and sorting these objects by its surface flatness. The main work of this thesis is as follows:(1) Analyzed the Three-dimensional vision technology and compared the characteristic and performance of various three-dimensional vision technologies. Especially, for this subject, a method for surface flatness sorting which used binocular stereo vision technology is proposed.(2) Research on the camera model to analyze the actual lens aberration, introduced the third-order radial lens distortion and tangential distortion items into the lens distortion model, which render the distortion model more consistent with reality. According to characteristics of the irregular-granular objects, based on the improvement of the image segmentation strategy, the calculation of matching cost, global evaluation function and the calculation of the final disparity map, proposed an improved image segmentation-based stereo matching algorithm.(3) Optimized and improved the stereo matching algorithm, made use of the power of GPU to accelerate the process through the technology of CUDA(Unified Computing Arch- itecture). So it can greatly reduce the running time.(4) Analyzed several eigenvalues which have a relationship with the flatness. The appropriate eigenvalues were determined through a large number of experiments. The results show that the method this thesis proposed is effective to solve the surface flatness automated sorting problem.