Research On Machine Vision And Uncalibrated Visual Servo In Micromanipulation System

The new technology is developing towards two extremes:on one hand, we can step out of the earth, and explore the deep universe tens of billions light years away; on the other hand, we can arrange a single atom, or build robots running in the human blood vessel. The micromanipulation is one of the forefront of micro-field research. It requires the high accuracy, repeatability and reliability especially in automatic and semi-automatic system, while the workspace is extreme small, the object is tiny and fragile, and the device is of weak precision. These extreme conditions make a mass of unique problems which are not in the conventional scale, and require the specular methods.This dissertation firstly reviews the advances of the visual servo, the micromanipulator and the machine vision. And then related modeling technology is discussed to provide a basis of theory for the following part.The objects in micro-vision have diffirent colors which can be used to distinguish those objects, but the external cold light source usually affects the true color. This dissertation proposed a density clustering method based on IHLS color space. The IHLS (Improved Hue-Lumination-Saturation) color space can solve the un-uniform problem of existing color space and reduce the influence of environment illumination. Then the improved DENCLUE method which is more fast and precise than the formers is used to segment the objects.The existing researches on micro-vision multi-objects recognition indiscriminately distinguish all the objects with one unified method, which is time-consuming and inaccurate. In this dissertation, the piezoelectric gripper, vacumn gripper, hohlraum and capsle, which are used in normal ICF (Inertial Confinement Fusion) assembly experiment, are recognized with difirrent algorithem respectively. Those methods significantly decrease the time consuming and improve the performance.Visual servo, especially the uncalibrated visual servo, is one of the cutting-edge researches in robot field. The online estimation methods are not precise, while the offline learning methods are not appropriate for most natural environment. A novel approach based on online support vector regression algorithm is proposed, which overcomes the drawbacks and combines the virtues mentioned above.In the end, this dissertation designs an automatic micro-manipulation system consisting of high precision stage, high-resolution camera, piezoelectric gripper and vacuum gripper. A motion controller based4-level software framework is also proposed to improve the usability of update and maintenance. The performance and effectiveness are validated with the proposed methods in the ICF target aseembly experiment.