| As a commonly used measuring instrument,analog instruments are diffusely used in industrial production and other fields because of their simple structure and convenient maintenance.Compared with digital instruments,they are not subject to electromagnetic interference.However,these types of meters are mechanically structured,unable to provide an interface for digital information output,and their data cannot be directly collected by a computer.Therefore,manual meter reading methods are mostly used for the readings of such meters,which is low in efficiency and subjective,especially in the face of real-time feedback,it is tough for humans to make timely and precise readings,making analog meters unable to adapt to modern management needs.Therefore,a new,reliable and efficient automatic meter reading technology is urgently needed.The main work of this paper is summarized as follows:1)Scheme design of meter reading system.In view of the intricacy of the image algorithm and the traditional meter reading and transmission of images,the system needs to be supported by a processor with great computing power and a network with large bandwidth.In this paper,Raspberry Pi is the core,equipped with a dedicated camera to constitute the entire terminal equipment,to support the computing power required by complex programs.The reading results obtained by image processing are transmitted via the network to the server for storage and forwarding to the monitoring center to facilitate staff management,instead of transmitting images to the monitoring center for processing,thereby reducing bandwidth pressure.2)Dashboard feature recognition.In view of the current research on specific types of analog instruments,such as the linear distribution of scale lines,which must include arcs and other features,lack of compatibility,and some require manual calibration in advance,new dial positioning,pointer and scale line recognition are proposed.Program.Use FLD to detect line segments and initially identify the scale lines,locate the pointer rotation center and part of the scale lines,and then determine the dial area;combine projection,thinning,and line segment detection to reconstruct the pointer axis;combine the smallest rectangle and the positioned scale lines identify all scale lines and classify them;tests show that the algorithm locates the dial and recognizes the pointers and scale lines accurately.3)Meter reading algorithm.In view of the fact that the current meter scale value recognition mostly uses templates and structural methods,it lacks a certain generalization ability.It is proposed to use the main scale line to locate the scale value,and then identify the scale value by the convolutional neural network;combine the look-up table and the distance method to determine the number of the meter.Calculation.Tests show that the entire algorithm reduces manual intervention,improves the automation of meter readings,and has wider meter compatibility.The maximum reading error does not exceed 0.74%.The processing time of a single image remains stable at about 0.88 s,satisfying the needs of accuracy and high real-time.4)Automatic meter reading system software development.In Windows environment,the joint development of the multi-language algorithm is completed by adopting C++ language integrated with Open CV image processing library and Tensor Flow deep learning framework of Python language,and then with the help of Qt interface development framework,the algorithm is encapsulated into software and cross-platform joint cross-compilation and packaging is achieved.In this paper,we develop the terminal of analog instrument reading under Linux environment,the server software of data storing and transponding,and the monitoring center software of data monitoring and long-range operation under Windows operating system.Through testing,the three sets of software run stably,and reach the requirements of automatic meter reading system. |
PDF Full Text Request