Automatic Meter Recognition For Inspection Robots In Substation

At present,substation inspections are mainly based on manual inspections.Under the background of manual inspection with the disadvantages of high labor intensity,inefficiency and insecurity,inspection robots were emerged quickly.It has become a hot issue of current research.And automatic recognition of meter is one of the essential core functions for inspection robots.Firstly,the robots carry the camera to obtain the pointer meter pictures to be identified.Therefore,aiming at positioning and identification of pointer meter,the objection detection based on SSD algorithm is studied.SSD network structure is based on the VGG16 network.Compared with traditional detection methods using the last layer of network for feature detection,SSD algorithm uses multi-scale layers for detection.Consequently,the algorithm can detect targets of different scales more accurately.Then to deal with the poor detection on small targets of SSD300 network,the SSD400 network is proposed.SSD400 network improves the detection accuracy of small objections.Secondly,in the outdoor working environment of inspection robots,the images are preprocessed.Focusing on the problem of uneven illumination of meter images,several common algorithms for correction illumination are compared.A correction algorithm based on two-dimensional gamma function is adopted.After extracting the reflection component,the brightness of images is adaptively corrected according to the light distribution characteristics.And the experiment is carried out on the value of the coefficient of the gamma function.The appropriate coefficient value is selected,which improved the effect of correcting the illumination.Lastly,we propose a pointer meter identification method based on maximum stable extreme value region(MSER)to extract the pointer region.The pointer region is extracted through two MSER detections.On the condition that the pointer passed through the axis of dial,the pointer is precisely positioned by thinning algorithm and progressive probabilistic hough transform.Finally,the coordinate is established and the number is represented by the pointer deflection angle.