Research On Automatic Reading Technology Of Electric Power Inspection Instrument Based On Computer Vision

Transformer substation is an important part of the national grid.Maintaining thenormal operation of the transformer substation requires routine inspection of the instrument,recording the relevant data and responding to the warning messages.Traditional manual inspection way is inefficient,personal errors and security risks existing in it.With the development of robot technologies and computer vision and other related technologies,it becomes an inevitable trend that the robots replace human to carry out inspection task of transformer substation.One of the core technologies of transformer substation inspection robot is visual processing part,namely localization and identification of gauges.Instrument positioning is to find the instrument position in the image to be recognized in order to extract the image of the dial area.Pointer meter reading recognition is to obtain the meter reading by identifying the pointer angle or other methods.For the positioning of the instrument,since the substation belongs to an artificial scene,there are many similarities,and the scene is complicated,so the traditional way of using features is easy to identify many similar features,so it is challenging to locate the position of the instrument accurately.Through the analysis of the inspection task,this thesis believes that the image in the electric power inspection has specific characteristics,that is,visible light intensity changes,rigid targets,and stable spatial relationships between objects.In this thesis,from the perspective of the human cognitive process,combined with the idea of target detection and image alignment,an instrument positioning method based on correlation filtering is designed.The image pre-captured by the robot is used as a template image,and the position information of the target instrument is manually obtained from it as a training sample to train the relevant filter.In this thesis,the relevant filter is transformed into a linear regression problem.Considering the obvious change of illumination,HOG features with good robustness to illumination are used as feature descriptors.The response map is calculated by calculating the correlation between the test image and the target area,where the area with the largest response value is the target instrument position.For the identification of gauges,this thesis simulates human vision mechanism and designs an algorithm which uses multi-angle Gabor filter.Because Gabor filter is sensitive to measure and direction,this algorithm uses muti-angle Gabor filter to filter the image,get the statistics of results and find the max result.Then the angle of the max result is the angle of pointer.According to the experimental results,the method proposed in this thesis has apparent advantages in terms of the robustness,real-time performance,and accuracy of meter positioning compared with the existing power inspection methods.At the same time,it can avoid the interference of factors such as shadows and complicated dials to achieve accurate recognition of the pointer reading.