| The operation and maintenance of transformer substation is a cumbersome,boring and repetitive task.The realization of comprehensive intelligent operation and maintenance of transformer substations by inspection robots has become the tendency of smart grids in China.And the research of automatic meter rcognition is one of the core part of intelligent inspection robot.Usually,the images acquired by the robot will be exposed to many effects,such as different poses,the illumination variations,object shelter and clutter background.In addition,there are many other challenges about instrumentation devices in the power system,such as various types,large number of devices,aging and damage of equipment.Moreover,the power system requires high accuracy for meter reading recognition.These problems pose challenges to the accuracy,speed and robustness of visual algorithms.Based on the above challenges,the research in this paper is based on the automatic recognition method of meter reading in substation scene,including the research on instrument object detection,image matching,pointer line positioning and meter reading recognition.The main contributions of this paper are as follows.(1)A keypoints location network(Pointer-Pose Net)based on convolutional neural network is proposed,and the network is used to locate the keypoints of the pointer line to obtain the pose of the meter pointer line.The use of keypoints positioning for instrument detection improves the accuracy and versatility of the pointer line positioning.And it can detect the pointer line of different types pointer meters accurately with a high speed.(2)A general method for automatic reading recognition of pointer meter based on template matching is proposed.This method is suitable for the recognition of all types about general-purpose meter in substation,and it has good robustness under obvious illumination changes and various fuzzy conditions.A large number of experiments prove that the accuracy of this method can reach 98%. |
PDF Full Text Request