Research On Machine Vision Alignment System And Development Of Related Technology

The structure of microelectronics industry equipment (such as automatic chip mounter) is complex. The vision automatic alignment system is one of the key parts of this kind of equipment. At present, the machine vision alignment system is generally consisted of special hardware and software, however the hardware is expensive and the software is hard to program. It is necessary to develop a system with common hardware and software that will shorten the time of system structure design, reduce the cost and easily program the software. This dissertation carries out systematic studies on the structure design of machine vision alignment system with high precision for automatic chip mounter, the development of related software system modules for computer control, and image processing and pattern matching algorithms.The vision alignment system, which is a special product, includes optic structure design and image processing algorithms. In this dissertation, the principle and structure of the automatic chip mounter with high precision are analyzed, the hardware structure of the optic alignment, which includes the vertical and horizontal direction alignment of bonding between a chip and a substrate, is designed to achieve sub-pixel accuracy. And vertical direction alignment includes two aspects: image alignment before bonding and the image refinement after bonding.Image processing technology is one of the most important technologies of vision alignment. Image preprocessing methods will affect the image localization precision. Image preprocessing algorithms necessary to pattern localization are detailed, and the image detection methods such as region of interest (ROI), threshold segmentation, edge extraction are deeply discussed.A new algorithm is proposed to resolve the localization problem in machine vision system. The wavelet feature coefficients are extracted by image transformation. RBF search process, which is based on radial basis function (RBF) principle, results in pattern localization. Experimental results show that the algorithm is robust against noises and not sensitive to rotation, translation, and scaling transform of image, and that sub-pixel localization accuracy can be achieved. At the end, the software system design and development are introduced, and all kinds of image algorithms discussed above are integrated in the software.