Research On Key Technologies Of Machine Vision Inspection System

With the development of industrial automation, the modern manufacturing and processing technology is improved continuously, and the consumer’s demand for product quality and consistency is becoming higher and higher. As an important part of product quality control, the inspection occupies an important position in production process. Because of high precision and non contact property, machine vision inspection technology receives attention and recognition in a wide range of industries. According to main factors affecting detection precision, in this thesis, the key techonologies of machine vision inspection system are in depth studied, especially for line scan camera calibration, radiometric calibration, subpixel precision edge detection and template matching.First of all, a new method of line scan camera calibration is put forword. On basis of the derived line scan camera imaging model, the projection matrix of line scan camera is solved by using least squares and internal and external parameters are resolved according to projection matrix’s characteristics. In addition, lens distortion model of line scan camera is studied, and the method of improving the calibration results is discussed by introducing the lens distortion model. Finally, the experiment proves the feasibility and validity of the calibration method.Secondly, the methods of radiometric calibration based on image sequence and based on a single image are studied, and these two methods are analyzed qualitatively on the basis of their experiments results. According to the designed experiment based on2D calibration target, the measurement results of machine part before and after radiometric calibration are compared. The experiments show that the radiometric calibration is of great significance to improve the measurement precision.Thirdly, a new method of subpixel accurate edge detection based on the improved Canny operator is proposed. On basis of definition of image edge, the basic principle and implementation of Canny operator are mainly studied, and the improved Canny operator based on adaptive threshold is proposed. Besides, the subpixel edge detection based on amplitude fitting method is studied, and the method that based on the variance of edge position of computing the adaptive neighborhood is proposed.Finally, according to basic principles of evaluating the template matching algorithm, the characteristics and performance that the superior template matching algorithm should possess is analyzed. In addition, the theory and principle of typical template matching algorithm is studied, and the method to improve the matching accuracy and efficiency is discussed further. At last, the stability, reliability and elapsed time of matching algorithm are analyzed according to the experiment results.