Design And Research Of Switchgear Monitoring Node Based On Raspberry Pi

At present,most traditional meter readings are done manually,and it is inevitable that they will be misread or misread due to the external environment.With the development of industrial vision technology,machines can replace manual labor for high-risk or complex tasks,this approach not only solves the manpower shortage problem,but also improved work efficiency.This paper applies image processing to embedded technology,researches the detection methods of meters and indicators,and uses Raspberry Pi 3B + as the hardware,the use of QT software to write a graphical interface and realizes visual inspection about the switchgear monitoring node under the embedded platform.The research content mainly includes the detection of indicator lights,pointer meters,digital meters,and testing of recognition algorithms on the Raspberry Pi.An indicator detection algorithm the combination of color salient features and contour features is designed.First,the image in RGB space is converted to HSV space,and then the red and green indicator lights are extracted.After dividing the contour of the indicator,and get the center of gravity of the indicator,judge the ON or OFF state,and finally recognize the indicator by combining the detection results.A pointer instrument recognition algorithm based on Line Segment Detector(LSD)algorithm is used.First,the pointer instrument is subjected to tilt correction processing.The binary map is processed using the Zhang-Suen thinning algorithm,Then,using the Hough transform and the LSD algorithm to detect the pointers separately,the speed and accuracy of the algorithm are compared,it was decided to use the LSD algorithm to detect the pointers.and finally complete the identification of the pointer instrument according to the angle relationship.A support vector machine(SVM)digital instrument recognition algorithm based on color segmentation and Histogram of Oriented Gradient(HOG)feature is proposed.First extract the red digital feature information in HSV space,and then divide it into single characters by horizontal and vertical projection method.Next,the normalized numbers are extracted using the HOG algorithm,and the dimensionality reduction is performed using Principal Components Analysis(PCA).The extracted feature vectors were input to the support vector machine of Grid Search(GS)for recognition,and the performance was compared with the threading method,KNearest Neighbor(KNN)and SVM respectively.The HOG-GSSVM algorithm has an average recognition rate of more than 96%,which meets the experimental requirements.The human-computer interaction interface was designed on the Raspberry Pi,and the recognition algorithm was embedded in the Raspberry Pi for algorithm performance testing.After that,the system design and application of visual monitoring nodes based on Raspberry Pi were completed.