| At present,the collection of meter readings in the subway power distribution room in China is still done manually by the staff,which leads to many problems such as low efficiency,missed inspection and difficult management of false inspection and greatly restricts the modernization construction of the subway industry.Therefore,the traditional working mode of manual transcription of instrument readings can no longer meet the requirements of modern safe and stable operation of subway.Intelligent collection of instrument readings has become an inevitable development trend.With the rapid progress of artificial intelligence,machine vision and digital image processing technology,more and more intelligent meter reading collection and inspection schemes are applied to substations,distribution rooms and other scenes,however,in the environment of subway power distribution room,there are still many challenges,such as the irregular shaking of instrument image acquisition equipment caused by subway vibration,which affects the image quality,multiple instruments need to be collected separately at preset points,and the difficulty in recognizing the reading of some pointer instrument structures.The main work of the paper is as follows:(1)A video jitter detection algorithm based on Siamese network is proposed.First use Phash algorithm for video under the stable state of two successive frames to simple sample,need input samples to the fixed size for the limited use of SPP can optimize the network makes the sample arbitrary dimension input maximum retained the original image information,through the network to extract the characteristics of the movement between adjacent two frames after vector using logistic regression as a classifier to complete sample input jitter or steady state secondary classification evaluation,with 93.7% accuracy.(2)The method of monocular vision positioning and ranging is used to collect multiple instruments at preset points.Determine template matching based on correlation coefficient of gray level as instrument detection algorithm to calculate the instrument center coordinates in the image,at the and according to the linear camera imaging model to calculate the relative position of the instrument connected to a video camera coordinate,and distributed according to the coordinates of the corresponding motion control commands collected in the image center,suitable size and no tilt meter image,acquisition success rate higher than 97.6%.(3)Aiming at the problem that pointer in complex pointer instrument is difficult to detect directly,combined with the characteristics of dial defect caused by pointer endpoint in binary image,an indirect identification method of pointer instrument reading based on Blob analysis is proposed.Firstly,R channel is used as gray image,adaptive median filter is used for image noise reduction,gray linear transformation is used for image contrast enhancement,and OTSU algorithm is used for image binarization to complete image preprocessing.Based on run-length coding mode to the binary image to form a Blob connected domain and segment the dial and dial the image difference method and least square method to determine pointer endpoint and coordinate to complete the pointer indirect detection instrument center,finally through the pointer rotation Angle and the full scale reading to identify a priori information is complete,average error is 2.1%.(4)Finally,the intelligent instrument reading collect system is realized.The overall structure of the system and the communication architecture between the hardware system and the software system are determined.The selection of each component is completed in the hardware system,and the overall workflow and the realization mode of each module are illustrated in the software system. |
