Generalized Hough Transform Based Image Matching Technology For RFID Chip Bonder

As RFID technology evolves, its application to national defense, industry and people's livelihood is of great prospect. At present, the main domestic types of RFID packaging equipment and other IC packaging apparatus needs high-speed and high-accuracy visual software systems as the major auxiliary tools. The accessible visual products in the market, however, are products that mainly developed by large foreign commercial visual software companies, such as MIL, Congex, HexSight and Halcon, etc. These products are convenient and highly efficient, while at the same time their costs are rather expensive. So developing a high-performance vision software system is necessary. This paper mainly studies the image matching algorithm based on generalized Hough transform according to the visual orientation characteristics of bounder modules in RFID packaging devices.The conventional generalized Hough transform (GHT) algorithm to describe images of arbitrary shapes is very effective, but it is limited to a single complex graphics. In order to identify specific target object from more complex graphics, this paper carries on some improvements based on the traditional GHT algorithm:1) A new way of thinking is adopted in the edge detection and boundary point gradient extraction. Unlike previous usage of Sobel or Canny in the image boundary extraction or other edge detection operators, a simpler and more convenient method is used: extracting image's edge according to the image binarization and corrosion technology, the result of which is more conducive to extract the image's shape information.2) Considering the limitation of the traditional generalized Hough transform's inappropriateness in application to identify specific target object from complex graphics collection, this paper adds a new idea of edge extraction. First, the edge of the template image and the target object are detected, and then the image is divided into sub-divided areas according to connected region, and an R-Table is established for each sub-image respectively. In the visual orientation, the sub-divided areas are polled, which can not only restrict the cumulative voting in image matching when using Hough transform algorithm, but also can filter some "interference" regions unrelated to the final localization result. Thus it can improve the algorithm's accuracy, while saving computation time as well. In the ending of this paper, performance indicators of the improved GHT algorithm like the visual orientation accuracy, algorithm robustness and processing speed are tested and analyzed through data test and comparison experiments.