Method for panoramic image stitching of nuclear fuel assembly based on key point detection network

This paper belongs to the technical field related to video image processing of nuclear fuel assemblies, and describes a panoramic stitching method for vertical motion videos of nuclear fuel assemblies. The process of this method includes: using a key point detection network to identify the upper nozzle of the nuclear fuel assembly to determine whether the fuel assembly image is tilted; if the image is tilted, performing image rectification work; subsequently using an object detection network to identify the grids of the nuclear fuel assembly; calculating the motion velocity of the grids according to the positional changes of the grids and the variation of frame numbers; estimating the velocity of fuel rods based on the grid velocity; converting the velocity into the motion distance between each frame; finally cropping each subsequent frame, sequentially stacking them downward into the panoramic image, and obtaining a complete panoramic image of the fuel assembly. This method overcomes the problem of panoramic stitching failure caused by the scarcity of fuel rod feature points, and has the advantages of a wide application range and the ability to perform automatic panoramic stitching in scenarios where fuel assembly images are tilted.