Mechanical Water Meter Reading Recognition System Based On Machine Vision

The mechanical water meter will calibrate the water meter before leaving the factory.At present,most of the water meter factories in our country are calibrating the water meter by manual meter reading,so that not only a lot of labor is needed,but also the accuracy of detection is greatly influenced by human factors.With the rapid development of science and technology,it is necessary to automatically identify the readings of the water meter to replace the artificial reading.This paper is to realize automatic recognition of water meter reading by machine vision technologyThe main contents of this paper are as follows:(1)The hardware and software design of the system:the design of hardware system is mainly set up the experimental platform,including the selection of camera lens and light source;the software system is the development of automatic recognition system of meter readings and opencv based on VC++ platform.(2)The recognition of pointer reading:the hardware platform is built up,collecting water pictures,the pictures obtained is RGB color model,in order to extract the red pointer,need to be converted to HSV color mode color extraction model,H channel,red is easily extracted;because in the picture there are more isolated noise points,so the need to remove the noise filter,most of the information but also to keep the picture,is used as the median filter for image processing image,the effect is very good;the image binarization processing,the pointer from the background highlighted.Then extract the pointer contour with the Canny operator,in the closing operation pointer optimization based on clustering contour;circle fitting method calculate the coordinates of the center through the pointer;distance method is used to calculate the pointer tip,with the coordinates of the center coordinates and calculate the finger tip can be calculated by the meter pointer to the deflection angle,and then according to the relationship between the pointer and shows the number of the realization of recognition pointer readings.(3)The word wheel digital reading recognition:the collected images to gray scale processing,not only can reduce the amount of data.Most of the information,but also can retain the image;Soble operator with the image edge detection,highlight the edge information of the image;using simple global thresholding for image binarization processing;line detection by Hough the positioning wheel frame,the wheel from further tilt correction,then according to the ratio of the known single character word width width of the frame wheel,divide a single character,and then the single character binarization processing,input to the support vector machine classification and recognition.The test shows that the system has strong real-time performance,high recognition rate and good adaptability.