Research On Recognition Method Of Pointer Type Meter Based On The Locating Scale Accurately

Substation is the bond between electric power and human beings.Whether the substation can operate normally without hidden dangers is the key to ensure the development of people's life and national economic.In the regular troubleshooting,the substation often relied on the method of manual inspection to record the equipment in the substation in the past.The way of manual inspection to record the equipment is going to cost high with low efficiency.Nowadays,robots gradually get into the power grid,and patrol robots replace the traditional manual work to take the equipment in the substation into pictures.When the patrol robots is working,the traditional reading recognition method is no longer suitable to recognize the pictures of oil-level taken by the patrol robots when the platform deflecting or the route of travel deviating.In this case,this paper aims to introduce a method of recognizing the readings of pointer type of oil-level meter based on locating the scales accurately.The main work of my thesis includes:(1)We propose a new method to improve traditional method of locating scales in instruments.When the we deals with the picture of the pointer oil-level meter whose scale is partially blocked due to the large shooting angle,traditional method can not locate scales of meters directly.In this thesis,I put forward an improved method,first,coarse recognizing all possible scale areas in the picture,and then selecting areas which contain the scales from the results of coarse recognizing by fine recognition,thereby accurately locate the scales of the instrument accurately.(2)A high quality data set of pointer oil-level meter is constructed.For the current situation that there is not only lacking the data set related to the oil level gauge,but also lacking hte data set of pointer type instrument with shooting angle,this thesis constructs a high-quality data set of pointer oil-level meter with shooting angle.(3)A method based on the deep learning of location scales of pointer type oil-level meter is proposed.Firstly,the numbers corresponding to all scales in the instrument are detected by using YOLOv3.Then,according to the center position of the circle fitted by the center of each Bounding Box of number,and through the relative position relationship among the center of the circle,the center of Bounding Box of number and the scale,the scale position is determined and positioned.This method can not only locate the scales information of the instrument,but also obtain the numerical information of the corresponding number of thescale,which is conducive to calculate the readings of oil-level meter.(4)A method to calculate the reading of oil-level meter with shooting angle is proposed.Aiming at the problem that the large shooting angle will cause the deformation of the oil-level meter dial.This thesis combines angle method and distance method,first finding the projection point from the fitting center and the pointer of meter,and then this projection point is selected as a new center to calculate the readings of meter.In this way,even if the shooting angle is large and the deformation of dial,this method can still calculate the meter readings accurately.The experimental results show that the proposed method can significantly improve the recognition accuracy compared with the traditional angle method and the method of image correction.The thesis introduces the reading recognition method of the pointer oil-level meter,which does not need to correct picture in advance.It can detect the pointer and positioning of the scale accurately,and then use the improved method of recognition meter to calculate meter readings.Compared with the other methods of meter recognition,the method in this thesis has obvious advantages in recognition accuracy when processing the meter pictures with large shooting angle.