Research On Embedded Meter Reading Terminal Based On Machine Vision

At present, the record and charge management of the water consumption are all to depend on the manual methods for the majority of water conservancy departments. Some of new meter reading technologies have many merits, such as automatic reading meter, centralized management and so on, but also have many faults, such as high reconstructing cost, big accumulative error and so on. In view of that, the dissertation researches one kind of embedded meter reading terminal based on the machine vision. Image acquisition technology, image processing technology, pattern recognition technology and GPRS communication technology are researched systematically, and their applications in the terminal are analyzed. Main research achievement is as below:According to the performance that the terminal must reach, the architectural structure of the terminal is built. The hardware system of the terminal with the ARM7 controller as the core is realized. The dissertation has finished circuit design of periphery memory module, image acquisition module, GPRS communication module and power module. In research on image acquisition module, two design proposals have been given. They realize image data acquisition and transmission based on MCU and video memory and based on CPLD and SRAM respectively. Through the comparison between two plans, the design principle of terminal is explained.The software design has been finished. According to the functional request that the software design must reach, the architectural structure of the software is built. The programming of image acquisition module, data communication module and data storage module is analyzed.According to the special features of water meter dial's image, the pretreatment plan is researched. In study of image smoothing processing algorithm, one adaptive image-smoothing algorithm combining non-linear and linear filter is designed, which gives dual attention to smoothing effect and image detail reservation. In study of binarization algorithm, one improved Niblack method is researched, which gives dual attention to binarization effect and processing speed. In study of character thinning method, one fast shape preserving morphological thinning algorithm is used.According to printed digit's features, one three-layer-structure recognizer is built. The recognizer takes the syntax parsing method, the template match method and neural network method in turn and realizes printed digit recognition with high efficiency and high accuracy. In the first layer, the syntax parsing method is used to make the simple classification;In the second layer, the template match method is used to make the precise classification based on the image grid characteristic;For the images that can't be recognized, the terminal sends them to the remote center, which uses neural network method to make the final recognition.After tested by practice, the terminal has met the technical request of remote meter reading system, and has laid the solid foundation for the water resources information management system.