The Key Technology Of Substation Maintenance Area Monitoring Based On Machine Vision

As a pivotal part of the power system,the maintenance of transformer substation is essential for substation’s normal operation.Under normal circumstances,critical electrical equipment,such as transformer and mutual inductor,must be taken out of service for maintenance or replacement after serving a certain time.In some emergent events,the staff must enter the substation promptly for emergency repairs.If the staff entered mistakenly,it may cause casualties and equipment damage as surroundings are high-voltage operating electrical equipment.In order to solve above problems,this study adopts stereo vision technology,on-site camera calibration,video image preprocessing,stereo matching and three-dimensional reconstruction to perform real-time positioning of maintenance personnel and reconstruct the three-dimensional model of the staff and electrical equipment in the substation maintenance area.The three-dimensional model and real-time positioning enable the inspector to monitor the movement of the staff clearly,and precisely measure the relative distance between the staff and the electrical equipment,thereby improve the safety of the staff.This research was conducted mainly on the following aspects:(1)According to the safety distance among the electrical equipment,the actual operable space,the coverage of the camera’s view and the factors such as the proportion and position of the measured object in the two camera images,the optimal calibration distance was calculated and the circular calibration plate was improved,in order to improve the image of the calibration plate,making it suitable for on-site positioning of the substation.An improved Hough transform algorithm based on filter and the least squares optimization was proposed.On the basis of standard Hough transform,the proposed algorithm completed denoising by adding different weighting window filter.Then the least squares fitting data was utilized to improve the positioning accuracy.Finally a strong feature point’s extraction algorithm based on Hough matrix peak was proposed to avoid the error and miss of matched pixels.(2)In the image preprocessing step,the commonly used object detection algorithm has several issues,such as poor background removal and incomplete retention of target information.In order to overcome the disadvantages of the commonly used object detection algorithm,a multi-frames target detection algorithm based on time domain,space domain and frequency domain fusion(TSFF)was proposed.The algorithm first selects the pixel values at the same position of the multi-frame images to form a set of n-dimensional vectors(time domain).Then the n-dimensional vectors were frequency analyzed by short-time Fourier transform(frequency domain).In addition,the pixel points of the adjacent area from the horizontal and vertical directions of the target were also analyzed.The mean value of the standard deviation of each group of time-frequency signals was calculated.In the end,double thresholds were set up to complete the target detection.The experiments showed that the TSFF algorithm performed better than other common algorithms in terms of background denoising and target retention.(3)Compared with the traditional Canny operator disadvantages of denoising and edge connection,this study proposed a new Canny operator based on the improved morphological filter and path predicting(IMPP-Canny).Firstly,the improved multidimensional structural element morphological filtering is used to replace the Gaussian filtering,which requires manually setting of the variance.The replacement can eliminate the artificial interference.Then a path prediction edge connection algorithm is proposed.The three corresponding to the current path direction are selected and the direction with the highest membership degree is used as the extended path to complete the edge connection.Experiments show that the IMPP-Canny algorithm has better denoising effect than the Canny algorithm,and it can fill the small hollow inside the target and connect the edge bands with larger connection gaps better,the connection path is closer to the real edge.(4)The commonly used matching method is insufficient to meet the positioning requirements of substation field staff and(or)has mismatching situations that will affect the positioning accuracy.In order to improve the situation,an algorithmthat combined an A-KAZE and improved SURF fusion operator(AKISF)was proposed.Firstly,the left and right image feature points were located through A-KAZE algorithm.By utilizing the nonlinear diffusion filtering equation,the image pyramid was constructed.And the fast explicit diffusive(FED)was used to speed up the operation.By this way,more details of the feature points will be obtained.Then compute the second-order response of image based on second-order multi-scale I-SURF vector to improve the quantity and quality of image.At the feature points matching stage,more matching points were selected by high threshold of BBF algorithm,and false matching points were eliminated by RANSAC algorithm.The AKISF operator was proved to have performed well in the number and accuracy of matching points,and can meet the site requirements.(5)Aiming at the large size of the substation electrical equipment,the relatively simple structure,and the difficulty of segmenting static targets from complex scenes,a binocular stereoscopic multi-angle fusion reconstructionalgorithm combining ICP and SFM(ISMF)is adopted.The ISMF algorithm first uses the SFM algorithm to capture images at four angles around the target through a binocular camera,manually selects feature points at the critical position of the electrical device and calculates the threedimensional coordinates.Then ICP algorithm is used to obtain the rotation matrix and translation vector of each set of adjacent points.Finally,point cloud fusion and 3D reconstruction were completed.Experiments show that the ISMF algorithm removes mismatched points and other unnecessary scene information,and the time is short,and the reconstruction results and positioning accuracy meet the site requirements.