Study On The Methods Of Vision Measurement And Reconstruction For Mini(Small) Objects

With the rapid development of computer technology and cost reduction of vision sensors, vision measurement and reconstruction technique have played an important role in the manufacturing and equipment industry based on its’ advantage of non-contact, high-precision, real-time.In recent years, with the improvements of the manufacturing methods and process,there has been an increasing number of micro (small) type products,(size2mm to100mm), and the corresponding visual measurement tasks are gradually increasing and put forward some new requirements, such as high precision, speed, and low false rate.This thesis focuses on the difficulties in the vision measurement and reconstruction of the micro (small) type object. Several miniature objects (such as electronic connectors, medical dental model, medical syringes, jewelry, small non-standard parts etc.) had been taken as the research object from2/3-D aspects. Some specific issues have been studied in this thesis, such as:the geometric characteristics of the micro and small objects extraction and recognition, the rapid reconstruction of precision shape, three-dimensional characteristics measurement, model simplification. Major works and innovations of this thesis include:1. For the vision measurement of the2-D geometry of the micro (small) type electronic products, a robust algorithm based on least squares and weighting function had been designed to extract the geometry features (line, circle and ellipse), meanwhile these features are taken as the skeleton base to locate the defects around.2. For the feature recognition of the micro (small) type object, a common affine invariant feature recognition algorithm had been designed, which utilized pseudo-Zernike moment and SIFT matching to solve the identification and classification problem when the features had been rotated, zoomed and transferred.3. For the fast reconstruction needs of the convex micro (small) type objects with less texture and rule shapes, an algorithm based on multi-view closed contours (Shape from closed contours, SFCC) had been proposed. The precision mesh model of the target can be obtained within a very short time (<40sec). In this algorithm, a sequence of contour was extracted from multi-view images, and then two-steps cutting was implemented to restore the surface model directly. In order to implement this algorithm, a serial hardware was designed and equipped. The diamonds are taken as the experimental objects. In the experiments, SFCC method can restore the3-D mesh quickly, and the distance precision can reach0.3(mm), the angle precision can reach0.03°, which can be used for the cutting grade for the diamond object.4. For the needs of complex irregular micro (small) type surface point cloud registration, a register algorithm based on the rotating platform was proposed. In this algorithm, a cylinder constraint was used to calibrate the position of the rotating platform in the space coordinate. Then any point cloud on the rotating platform can be registered automatic using the rotate angle information, without manual post-processing or sign point constraint. This algorithm can be combined with many types of3-D scanner. Experiments show that the precision is the same as the leading point registration algorithm, which is less than0.3(mm).5. Some geometric features (such as plane, sphere, and cylinder) had been extracted based on point cloud. Flatness was calculated using the plane feature. Sphere and cylindrical features of a non-standard work piece was extracted using related methods.6. A simplification algorithm using multi-directional cross-section lines based on triangular data was proposed. This algorithm can generate simplified data efficiently and output quadrilateral and triangular grid data. Through comparison with the original triangular data, the simplified error was less than0.04(mm), while the compress ratio can be under10%. This algorithm has been used for the oral disciplines dental wax stress analysis.