Direct-Reading Instrument Automatic Calibration Method Based On Computer Vision

Periodical calibration is important technical operation for making instruments ingood condition. In traditional calibration method, the needle position relative to dial isdetermined by manipulator vision, and lead to tested accuracy is influenced by man'sjudgment. The calibrations accurate and fast are most critical requirements of thecalibration phase. In this paper, a new measurement instruments calibration method isintroduced based on computer vision mechanics. The fixing viewpoint imaging model isdeduced. The needle and scale calculating method is optimized, and the dial elements areclassified. Finally, the measurement errors are analyzed, and the foreground isrepresented in the power system. The theory and implementation technique is studied,and the achievements and results are included:The new calibration approach based on computer vision, which using visual imageanalysis extract the needle and dial features, and the recognition results is utilized tocontrol the calibration process instead of operator eyesight. The scheme is belong tocalibrate the instruments using standard power source. And it is composed ofprogrammed power source, digital camera, image acquired board and computer. It is aclosed-loop control system, and it integrates the optic, electric and computer, completethe instrument driving, reading and error analyzing.Aimed at the moving viewpoint based on the instrument reading rule, the imagingmodel is analyzed utilize the imaging theory;the new algorithm is deduced at the fixingviewpoint satisfied the instrument reading rule, it can reduce the deflection in theinstrument reading.The light reflection of the meter glass-cover is studied and its category isdetermined, and the homomorphic filter is utilized to reduce interferes. The segmentmethods are discussed, and the best threshold is determined using histogram and mostvariant-ratio method.Based on projection and kerf technology, the pointer and scales area is locationaccording to the edges of meter image. The line parameter estimation utilized the leastsquare and point to arc method is represented. The point to arc algorithm can determinedmultiple lines at meantime, and can discriminated the long and short scales. Itinsensitivity to discontinues line and acnode.In the procession of power meter automatic recognition, three particle featureinvariants, include length ratio, compactness and simplicity factor, have been extracted.The RBFNN is utilized in the dial plate elements recognition. The meter identifyingexperiments proved that the RBFNN input invariants are stable and proper, and it isinsensitive to the background noise. Not only it can be utilized as important reference tocalibration other kinds of instrument, and it also can optimize the power system dynamicparameter acquiring and running state monitoring with the image-analyzing introducing.