Research On Key Technologies Of Meter Image Recognition

Recognition system of meter reading is a kind of automatic examination system, which integrates optics, mechanics and electricity.Without changing original instrumental equipment, one meter reading image is photographed by machine vision technology, then the object of photograph is extracted and recognised in order to obtain needful information of meter reading which can be exported through standard electron ports. This system can easily realize that all kinds of meters with or without digital output ports constitute an integrated and automatic examination system.The recognition and processing system can improve automatization, communication competence and network connections competence of examination system in the industry automatic control domain. The recognition and processing system is involved in many research domains, such as machine vision, image engineering, artificial intelligence, robotics motion control, micro-electromechanical system and electrical driver, meter examination. There is very significant meaning that the technologies of meter image recognition are researched. The paper only researches the key technologies of meter reading, and doesn't carefully discuss other conventional technologies.The primary results and innovative points of this dissertation are summarized as following:1. Weighted recursive approaching filter is adopted to filter meter images.By changing attenuation coefficientα, the filter can obtain a good filtering output which other filters can't obtain. Ridgelet transformation is used to extract line-like characteristics such as blurred scales, pointer, in meter images for the first time.The ridgelet transform can take out beeline feature and the result is favorable.2. Support vector machine and invariant moments are used to recognise the symbols of meter types and precision ratings.Mobile, rotated, zoomed, rotated and zoomed symbols of meter types and precision ratings are recognized respectively.Experimental results indicate that the method is effective.Contrastive experimental results show that the recognition rate and recognition time of support vector machine which is based on the osu_svm3.00 toolbox are superior to those of BP neural network.3 For the sake of diminishing collection error of meter image, the right position of camera lenses is analyzed in the common dial meter reading recognition method. The arithmetic expression of importing error is deduced when the optical center of camera lenses is not at the top of the meter axis. The importing error is simulated and calculated according to the arithmetic expression. The servo tracking reading method and the calibration reading method are used to recognise high-precision meter reading respectively. The experimental results show the two reading methods are applicable to automatic reading of high-precision meter.4. The dissertation adopts image processing method to automatically measure damping time of dial meter for the first time.The new method can measure damping period of dial meter rapidly and accurately.