| Machine vision technology has the advantages of non-contact, high precision and rapidity. Industry leaders begin turning their attention to machine vision-based test automation and assembly automation, after solving the problem of production automation. Machine vision technology has become one of interest topics in industry.In the printed circuit board manufacturing process, the positioning accuracy of location hole, directly determines positioning accuracy and alignment accuracy of PCB in the following processes in the circuit board, thereby it can affect the quality of PCB. Location hole is positioned and main axis of CNC system is guided to align with the location hole by machine vision system. It can effectively improve the hole positioning accuracy, and can significantly increase production efficiency.To achieve the size of appearance and features of PCB, edge detection theories commonly used are studied and implemented. Their advantages and disadvantage are analysized by experiments. In order to overcome influence upon edge detection produced by the external illumination changes, the problem of edge dectection from image with noise is researched. A new fast edge detection method, which can suppress Gaussian white noise, is proposed.To improve detection accuracy, subpixel edge detection method is studied. Based on the fact that differential at neighbour of the edge point is a symmetric function, a common subpixel edge detection method is proved.In order to extract features on PCB, first of all, some corner detection operators including Moravec operator, Harris and SUSAN operator are analyzed. For solving the lack that SUSAN algorithm is not good in detecting X corner, by defining a concept of point pair, A novelty method for corner detection by matching point pair and nucleus is proposed, and experiments show that corner detection has better detection rate and higher robustness than SUSAN. Subsequently, the basic idea of Hough Transform and its various derivative methods are studied. Extraction methods of Various Geometrical features are analyzed, such as straight line, circle, ellipse, etc. For a straight line features extraction, a new method of straight line extraction based on first-order moment is given; For Circle feature extraction, according to the geometric characteristics of circle, a new method is given, which have low computation cost. In order to enhance the circle and ellipse detection and positioning accuracy, the nonlinear optimization formula of the parameters of circle is deduced, so the methods have the sub-pixel accuracy.To measure size of features and position, linear model and non-linear model of camera are studied. Based on these models, two kinds of camera calibration methods are analyzed: a method based on radial constraint and a coplanar calibration method. The process of solving inside and outside parameters of the camera model is given in detail. To further improve the calibration accuracy, nonlinear optimization methods are discussed. L-M optimization algorithm is used to optimize the initial the parameters, and the calibration accuracy can be improved.In order to research and validate technologies related to machine vision detection and positioning system, this paper designs a test platform with four degree of freedom. It can move by X, Y, Z directions and rotate around the Z-axis. A control circuit is designed using F2812 processor as the core device. A decoder circuit is designed using CPLD as the core device, and it can decode input signals of 4 channel encoders. In VC development environment, we design a set of image process software with the above-mentioned algorithms. In CCS development environment, the DSP control procedure is developed. Finally, they are organically integrated and built into a complete vision inspection and positioning system.Finally, experiments show the system can carry out the tasks: measurement, positioning and orientation adjustment, target tracking. Functions and key algorithms of system are validated.
|
PDF Full Text Request