Recognition Technology Research Of Meter Reading Based On Machine Vision

In the current meter calibration and manual meter reading process, often using artificial interpretation, not only is a huge workload, but also not easy to manage, so it is necessary to develop a rapid, accurate and stable automatic meter test. With the continuous development of computer technology and machine vision technology, the automatic recognition of meter reading becomes possible. To realize intelligent processing with the meter reading, this paper uses image processing technology to extract and detect automatically the pointers and digital display area of meter, then adopt an effective algorithm to recognition the meter reading.The main work of this paper includes:(1)Target location: A dial coarse location method based on edge detection was used to locate the dial region. Firstly, Using Canny algorithm for image edge detection, and then morphological and Connectivity-Segmentation algorithm was used to remove interference to get the dial region. Finally,according to the structural characteristics of the region, get the center and radius of the dial by calculating, realized the dial region fast location.(2)Digit extraction and recognition: Using the method of line detection based on dial wheel frame correction, wheel location and dial correction at the same time. After wheel area’s coarse positioning, adopt Hough transform to detect wheel frame lines to get inclination angle, then the image rotation corrected by affine transformation, finally the integral projection method was used to get round of regional characters. In the reading of the whole character recognition, using the classification method based on skeleton, the method of template matching based on feature points was used to recognition semi character, according to identify the second half word length interception half word template, extraction thirteen statistical features of half word image and half word template image,calculating the Euclidean distance among the image feature vectors, the minimum distance is the template recognition result.(3)The detection and identification of pointer: Adopt color conversion method to extracting image,using the morphology remove the noise of pointer image. Calculated pointer rotation center according sub dial connected domain properties, using triangle template matching method detection direction indicator, at last, according to the associated template sequence number and pointer reading get the pointer reading. Finally, using GUI based on MATLAB programming, showing the process and results of the meter reading with the automatic recognition, verifing the correctness and feasibility of the algorithm.