| The pointer instrument has been widely used in the industrial production and metrology verification.The pointer instrument should be verified regularly for safety and quality of the production.Now,the machine vision has been introduced into pointer meter auto-reading,but present methods are mainly focused on low precision pointer meter auto-reading.For high precision pointer meter,these algorithms are inefficient and low-accuracy and cannot meet practical need.In this dissertation,pointer meter intelligent recognition technology was researched and main study object was precision pointer meter of 0.2 accuracy grade,and the pointer meter of division value of 0.001mm was chosen as the comparative object.The main studied problems include:scale region segmentation,pointer extraction,figures of scale value recognition,error correction and auto-reading method.The main contents are expressed as follows:1.To solve the problem of big size pointer meter,the image acquisition system was designed.The structure of image acquisition system was studied,and then hardware equipment was decided.Finally,illumination system was designed and high quality meter image can be acquired.2.Aiming at the problem of scale region accurate segmentation,scale region segmentation based on intersect cortical model(ICM)was proposed.The image was operated through dynamic thresholding method,then the binary image was taken as the input of ICM.Finally,weighted Otsu was combined to calculate adaptively ICM iterative times to eliminate noise around the scale.The scale can be segmented accurately,and the arc of the scale can be fitted for a polar coordinates transformation for further processes.3.To extract pointer and correct image projection error,image subtraction method based on projective transformation was proposed.Corner detection algorithm was used to detect feature points and matching algorithm was proposed to find all possible point pairs between template image and meter image.Then transformation matrix was generated from point pairs,and the projective transformation was applied to meter image.Finally,these two images were subtracted to find the pointer.And the features of scale value were designed and used in character recognition.Through these,the adaptivity of visual system can be enhanced.4.To correct the error of monocular imaging model,the error correction method of monocular imaging model for precision pointer meter was proposed.And the reading method with high accuracy was presented which makes use of intersection point position of the pointer and scale.Proposed methods and algorithms in the dissertation were tested in simulation model.The result of experiments shows that proposed methods can meet the practical use of pointer meter of 0.2 accuracy grade.The relative error between meter auto-readings and actual readings less than 0.15%. |
