Research On The Method Of Infrared Panoramic Image Mosaic

With the rapid development of computer vision and image processing technology,people's research on the image has expanded from visible images to thermal infrared images.The thermal infrared image reflects the temperature image of the object,which has a good ability of anti-jamming,discrimination and camouflage,and can work all-weather.Therefore,it is widely used in military and civil fields.However,the thermal infrared images generally have the disadvantages of low spatial resolution and narrow field of view.It is difficult to meet the requirements for those applications that need the wide field of view and high resolution panoramic images.Therefore image stitching technology has become a main trend for obtaining panoramic images,so it is of great value and significance to study the infrared image stitching method suitable for large scenes.This thesis focuses on the realization of panoramic image mosaic.In the pre-processing stage,an image enhancement method of self-defined convolution kernel is proposed to improve the quality of infrared images.In the image registration stage,the registration algorithm based on the Scale Invariant Feature Transform(SIFT)is deeply studied.Aiming at the problems of low registration rate and poor real-time performance of traditional SIFT algorithm,an improved SIFT registration method based on the region of interest(ROI)is proposed.Firstly,the ROI of adjacent images is calculated,then SIFT feature points are extracted from ROI and used as moving targets,and the Kanade-Lucas-Tomasi real-time tracking algorithm is used to complete the first matching.Finally,random sampling consistency(RANSAC)algorithm is used to filter to remove the invalid match pairs,the registration rate and running efficiency of the algorithm are improved obviously.In the image fusion stage,because the traditional fusion method is too simple to deal with the stitching problem,an improved optimal segmentation line fusion method is proposed.Firstly,gamma correction is used to improve the contrast and brightness of the registration image,and then the similarity of the registration image in ROI is used as the energy function of the texture difference.Finally,the neighborhood information value factor is introduced into the segmentation line search strategy,which makes the fusion algorithm more flexible and can realizes seamless and no ghosting splicing of multiple infrared images.In this thesis,multiple sets of infrared image sequences are selected to verify the above theory.The experimental results show that the algorithm proposed improves the registration rate by 3.491% and the running time by 54% compared with the traditional SIFT algorithm.Compared with the traditional fusion algorithm,the proposed fusion algorithm has significantly improved the human vision and objective evaluation indexes.In general,the proposed method can accurately and efficiently obtain an infrared panoramic image with rich information,wide field of view and high resolution.