Research On Reading Recognition Algorithm For Pointer-type Meter Based On Machine Vision

The pointer-type meter has the characteristics of simple structure and low cost,and the swing of the pointer indicates how the data changes visually and directly.It is widely used in factory instruments and vehicles.As the degree of automation increases,it will be an inevitable trend to recognize the reading of the pointer meter by means of machine vision.Therefore,it is of great theoretical and practical significance to study how to improve the accuracy,robustness and applicability of the pointer-type meter reading recognition algorithm.The existing algorithms still have some problems,in which the meter area cannot be well positioned,many parameters need to be selected according to the scene in use,and the projection of the pointer on the image may deviate from the actual position to cause reading error.In response to these problems,a complete pointer-type meter reading recognition algorithm is presented in this paper,including meter area positioning,meter preprocessing,pointer extraction and line detection,pointer tip repositioning and reading calculation.The main research contents of this paper are as follows:(1)The method of meter area,start and end scale positioning is studied.A meter reading recognition algorithm is proposed based on meter area,start and end scale positioning.Firstly,for the problem of low efficiency and poor robustness when locating the meter in a complex environment,the YOLOv3 model is improved for the self-built pointer meter dataset to accurately determine the position of the meter and the numbers while accurately classifying them.Then,by calculating the positional relationship between the position of numbers and the center of gravity of the main scale lines,the position of the start and end scales in the meter image of any posture can be accurately positioned to avoid the dependence on the manually scaled fixed scale position when reading.Finally,the meter reading is calculated based on the relationship between the start and end scales and the position of the pointer tip.(2)The meter preprocessing,pointer extraction and line detection methods are studied to obtain the desired pointer tip position for reading.Firstly,in the target area of the meter,the image is grayed out,denoised,and enhanced which will highlight the feature information of the meter pointer and scales.Then,according to classification result of the meter,the related parameters of pointer detection are set adaptively.The pointer region is extracted by threshold segmentation.Finally,the stable detection of the pointer line is realized by the Canny edge detection and Hough transform which is optimized by angle and parameter constraint.And,the position of the pointer tip can be determined.(3)The method of correcting the meter reading by repositioning the pointer tip is studied.In order to solve the error of the pointer-type meter reading caused by projection imaging,a pointer-type meter reading correction algorithm is proposed based on binocular stereo vision.Firstly,an image of the pointer-type meter image is acquired by using a calibrated binocular camera,and the left and right images are aligned in the horizontal direction by stereo correction.Then,the SIFT feature points in the number regions of the left and right images are extracted and the matching between the feature points are completed.Finally,the meter plane is constructed according to the threedimensional information of the three pairs of matching points,and the pointer tip is repositioned on the plane.The algorithm is tested by building a binocular vision system.The experimental results show that the algorithm can locate and classify the meter with a correct detection rate of 97.5%.On this basis,the pointer tip is extracted and the repositioning of the pointer tip is completed,which reduces the reading error caused by the projection imaging by more than 60% and realizes accurate recognition of the pointer-type meter reading.