Research On Reading Recognition Method Of Pointer Instrument Based On Substation Inspection Robot

As the terminal equipment of state data display,instrument has been widely applied in substation,which can effectively guarantee the safe and stable operation of substation.Among them,the pointer instrument is installed in large numbers in the substation because of its stable working performance under the conditions of bad weather and strong electromagnetic in the outdoor environment of the substation.The reading and transcribing of instrument reading in substation are usually completed by manual inspection,but this method has high economic cost and certain safety risks.In addition,the work of manual inspection is monotonous and repetitive,which relies on workers’ experience.Long time work is easy to cause eye fatigue and lead to misreading,which is not conducive to the improvement of management level.We analyze the problems and challenges faced by the automatic recognition of instrument reading,and puts forward the corresponding method of automatic recognition of instrument reading.In this paper,the problems of instrument detection and extraction under complex background,instrument tilt distortion and center line fitting of dial pointer are studied and analyzed,and the following solutions are adopted.(1)A YOLOv4 based pointer instrument detection platform was built.The YOLOV4 object detection method based on convolutional neural network is used to detect and extract the instrument.The pointer instrument detection model is trained through the instrument image data to realize the detection and extraction of the instrument area and improve the accuracy,speed and robustness of the instrument detection.At the same time,combined with mobile robot,network communication and other technologies,we integrate the corresponding program of the inspection robot with the instrument reading recognition program,build the software interface is written.Finally,the instrument reading recognition system of the substation inspection robot is established,which can achieve efficient human-computer interaction.(2)A QR code information fused instrument image correction and reading recognition method is proposed.In order to solving the problems of inconvenience of image correction due to the inconspicuous features of the dial in the tilt and distortion of the instrument image,and the difficulty of extracting the range information in the dial due to the numerous types and specifications of the instrument in the reading calculation.A method of pasting QR code over the instrument is proposed.We detect the edge line of QR code and its intersection points,which are used for perspective transformation.Then the instrument image is corrected.At the same time,the prior instrument information is obtained by identifying the QR code to assist the instrument reading recognition.(3)The improved method of accurate fitting and extraction of the center line of pointer based on RANSAC is adopted.The extracted pointer contour is easily affected by various identifiers in the dial,and outliers caused by identifiers will appear in the skeleton after the pointer contour is thinned,which will interfere with the fitting accuracy of the center line of the pointer,so the RANSAC(Random Sample Consensus)is adopted to eliminate outliers and fit the center line of the pointer accurately.Thus,the identification accuracy of the instrument reading is improved.