Research On Automatic Interpretation Method Of Pointer Meters Based On Computer Vision

Pointer meters has the advantages of simple structure,low price and strong antielectromagnetic interference ability,so it is widely used in industrial sites.At present,the pointer meter mostly adopts manual reading method,but the method is easily affected by the observation angle of the meter reader and the reading environment,resulting in low reading accuracy.Therefore,it is important to study accurate automatic reading methods for pointer meters.With the rapid development of artificial intelligence technology,it has become a potential solution to automatically identify the readings of pointer meters using computer vision technology.This paper takes the pointer pressure gauge as the research object,and studies the automatic interpretation method of pointer gauge representation based on computer vision.The main research contents are as follows :(1)To address the problem that most traditional pointer meters reading studies are based on pre-acquired high-quality images(meter target is located in the center of the image and the meter dial is parallel to the camera plane),Faster R-CNN(RegionConvolutional Neural Networks)and Mask R-CNN target detection algorithms are introduced to automatically detect the location of the pointer meter target envelope in the current field of view,which simplifies the equipment installation and calibration process before reading.(2)In response to the problems of tilt and rotation of the acquired pointer meter images due to the deviation of the shooting angle by manual or inspection robots,the SIFT(Scale-Invariant Feature Transform),RANSAC(Random Sample Consensus)algorithm and perspective transformation are used comprehensively to obtain images without tilt and rotation,overcoming the influence of tilt or rotation of the pointer meter on the accuracy of automatic readings.(3)Aiming at the problems of unstable detection and low precision when using Hough transform series algorithm to detect scale and pointer,a circular scale search algorithm is proposed.This method first finds the search range of the scale by the center position of the digital area corresponding to the upper limit and lower limit of the range of the pointer meter;then the circle is drawn with the center coordinates as the center and the distance from any digital center position to the center as the radius.By gradually expanding the search radius range,all points intersected with the scale line are searched,which effectively improves the stability and accuracy of scale detection.(4)In the pointer representation value acquisition method:(1)To improve the accuracy of the automatic reading method of the pointer meter,a progressive reading method is used.The coarse reading is obtained by first identifying the large scale of the pointer meter,and then the fine reading is obtained by using the angle method between two adjacent large scales of the pointer.(2)In order to overcome the shortcomings of the traditional angle and distance methods,a Convolutional Neural Networks(CNN)-based pointer meter reading method is proposed.Firstly,a circular scale search algorithm is used to find the nearest two small scale positions to the pointer;then the CNN is used to read between the two small scales adjacent to the pointer.The experimental results show that the maximum error of the reading of this paper’s method is all lower than 0.7%.