Research On Multi-Channel Video Surveillance And Panoramic Image Mosaic For Inner Wall Of Thread-Barrel

Thread-barrel is an important tool of material transmission channel in nuclear power plant,the integrity and cleanliness of the inner thread is critical to prevent nuclear leakage.In some nuclear power plants,a single camera performs a spiral descent to collect video for the inspection of defects in the thread through human observation.It is difficult to locate rust or other damage of thread accurately due to the strong subjectivity and visual fatigue.In order to improve efficiency and accuracy of detection,this paper takes advantage of machine vision technology to avoid false detection caused by visual fatigue and the hidden danger of nuclear radiation to workers.Considering the large size of thread barrel,image sequence could be collected by using multi-camera rotation a week in the thread barrel.In this paper,we studied the distortion correction preprocessing,horizontal registration,vertical registration and image fusion technology for thread image,and developed the corresponding multi-channel video surveillance and panoramic splicing system.Finally,the automatic testing of thread rust was carried out.The main contents of the paper are as follows.Firstly,since there existed radial distortion in original image,the cameras were calibrated and the camera parameters were obtained.Polynomial was used to approximate the radial distortion of imaging device,and the distortion parameters were obtained by multi-parameter cooperative adjustment.Then,the original image was corrected by camera parameters and distortion parameters to obtain distortion-free images effectively.The process of panorama stitching included two parts,i.e.horizontal stitching and longitudinal stitching to realize the stitching of multiple images at different rotation angles.Firstly,SIFT feature points were extracted based on calibration overlapping region of horizontal adjacent images,and the rough matching was obtained based on the FLANN.The horizontal matching region was determined by coordinate constraint and RANSAC purification.Secondly,Hough linear feature Extraction,linear screening and linear matching were used to determine the offset between the longitudinally adjacent thread images;then,a number of cameras longitudinal stitching could be implemented on the basis of horizontal stitching.Lastly,multi-resolution fusion and natural transition between images could be achieved based on local brightness adjustment and the best suture search.The system on multi-channel video surveillance and panoramic image mosaic splicing was designed and implemented based on C ++ language.Through the horizontal and vertical local splicing tests,the horizontal threading trend is natural,both the spacing and arrangement of longitudinal thread are uniform,which verifies the accuracy of the stitching algorithm.Through the whole panoramic splicing tests of the rust and clean screw barrel,the number of threads in both panoramas is exactly the same as that of 35 threads,and the rust can be accurately located,which verifies the robustness of the stitching algorithm.Compared with the traditional single camera helix descent monitoring scheme,the system could monitor the original time-consuming 1 minute down to 20 seconds,which could greatly improve the monitoring efficiency.What's more,thread rust degree can be detected by color histogram which shows the intelligence of the system.In addition,this system can be applied to other aspects of the ring device monitoring and panoramic mosaic.