基于虚拟红外双目视觉的GIS三维温度场重构OA

Infrared thermography is an important means of temperature rise detection for GIS(gas insulated switch-gear).Infrared thermography devices on site mostly use infrared or infrared-visible two-dimensional imaging meth-ods,which are insufficient for characterizing the spatial thermal distribution of GIS.In this paper,based on a monocu-lar infrared camera,an adversarial neural network(GAN)is used as a virtual infrared thermal imaging camera with lateral parallax,and then a 3D reconstruction of the GIS temperature field is achieved through physical and virtual in-frared binocular visual stereo matching.The 1017 pairs of original IR binocular dataset containing different angles is collected by performing temperature rise tests of GIS under different currents and using zhang's calibration method to calibrate the physical and virtual camera positions.The image collected by the monocular infrared camera is used as the left image input,and the pix2pixHD image generation frame is used as the virtual infrared camera to generate the right image.The SGM(semi-global matching)stereo matching algorithm is used to match the original image with the generated image and to obtain the computed parallax map,thus achieving 3D reconstruction by solving the 3D co-ordinates in accordance with the parallax map.The result shows that the FID(fréchet inception distance)of the im-age generated by pix2pixhd image conversion framework on the test set can reach 55.136,the perceptual loss can reach 0.079,the structure similarity index measure can reach 0.926 and the peak signal-to-noise ratio can reach 27.387.The virtual binocular camera can reach the accuracy of physical binocular camera,which provides an effec-tive method for achieving economical and efficient 3D temperature field reconstruction of GIS and improving the reli-ability of temperature inspection results.