Key Technology Research On Intelligent Checking Ammter System Based On Computer Vision

Intelligent checking ammeter model is needed urgently to displace artificial checking ammeter system in order to solve problems exist in artificial checking ammeter system in electric power management department. The emphases on this paper is research on ammeter readings recognition in intelligent checking ammeter system and the main research contents and innovation achievements are described as follows:Firstly, approximate location method on ammeter readings region based on color space projection is proposed. The Hough transform is used to detect the outside borders of ammeter and find the oblique angle, and then the image is rotated according to the oblique angle. Because of illumination, the image intensity is not uniform. The classic binarization algorithms such as Ostu algorithm and Bernsen algorithm can not exact object well, so we choose binarization algorithm based on image partition that derived from Da-Jin method. The method is proved to exact object well. At last, the location method based on histogram projection is used to locate ammeter readings region precisely.Secondly, on the base of research on projection method and iterative algorithm based on continuous subset, an improved method based on above methods is proposed to partition characters that are seriously adhesive or fractured exactly. The improved algorithm can resolve problems that projection method and iterative algorithm based on continuous subset can only partition the good quality image.At last, because template matching method and sliding template matching method can not process the inferior image, the recognition method based on BP neural network is chose to recognize the ammeter reading. The experiment results show that it is superior to above two kinds of template matching methods obviously. Especially on the condition of processing inferior image, the method show strong capability of half digit character recognition and occlusion character recognition. The whole recognition rate of clear image is up to 98.8 percent and that of inferior image is up to 93.3 percent. The experiment results show the recognition rate meets practical need.