Research On Automatic Recognition Of Analog Instrument Based On Inspection Robot And Computer Vision

Analog instruments of power devices are widely used in Chinese substations.To maintain normal status of power devices and prevent damages of the substation,it is significant to monitor the indication of analog instruments via power inspection.Nowadays in China,in tasks of power inspection,manual inspection is the mainstream approach.However,manual inspection is inefficient,costly,and error-prone,which may cause damages to the substation.Meanwhile,some proposed robot-based power inspection approaches of the analog instrument are of low universality and difficult to achieve.So based on the existing inspection robot platform EXJR-HUST/SN-1,a vision system of analog instrument recognition has been developed independently in this paper,while related algorithms has been also proposed.Firstly,by investigating and analyzing relevant research status at home and abroad,a vision system of analog instrument recognition has been designed for robot-based power inspection.This vision system takes Nvidia Jetson TX2 and STM32 as the processing units,and is integrated with a GigE camera,stepper motors of horizontal/vertical directions and related encoders of stepper motors.This vision system can adjust the posture of camera precisely via the close-loop control of the above stepper motors.Then,a novel approach of camera posture adjustment which is suitable for robot-based power inspection has been proposed to make the camera align to the tested analog instrument in the process of power inspection.In the first step of the above approach,the information of robot localization has been used to adjust camera posture roughly.In the second step of the above approach,a novel fast camera alignment algorithm which is based on Histogram Oriented Gradient(HOG)and Support Vector Machine(SVM)has been used to adjust camera posture precisely.The result is that camera can be aligned to the tested analog instrument precisely via the above approach.Finally,a novel analog instrument recognition algorithm which is suitable for robot-based power inspection has been proposed to obtain the indication of the tested analog instrument in the process of power inspection.Speeded Up Robust Features(SURF)have been used to calculate the mapping relationship between the actual image and the standard template image.The distortion caused by camera angle can be corrected by projecting the pointer pixels of the actual image to the standard template image,and precise recognition result of the tested analog instrument can be obtained by calculating on the above reconstructed image.Indicated by results of experiments,the above vision system of analog instrument recognition is robust to different brightness levels of the background and different height of the tested analog instrument,while there is an extraordinary performance in recognition precision and abnormal indication warning.