Research On Key Technology Of Infrared And Visible Image Fusion System Based On Airborne Photoelectric Platform

The airborne photoelectric platform with infrared and visible light sensors can work day and night,and the infrared and visible light sensor images contain a large amount of complementary information,which can improve the accuracy of UAV detection and identification.Aerial image imaging environment is complex,weather changes,lighting changes and UAV vibration and so will have an impact on the image quality,and the flight of the role of the distance is too high,resulting in airborne photoelectric platform to obtain the image of the air to ground target complex background,an increase of infrared and visible light Image fusion difficulty.Thus it is in urgent theoretical and practical needs to study how to effectively improve the target indication of fused images,eliminate the effects of noise and relative image deformation on fusion accuracy,and propose a fast high-precision registration method for different-source images and a fusion method with rich complementary information and strong noise resistance.This paper discusses the basic theory and dynamic development of infrared(IR)and visible image fusion,and systematically studies every link of such image fusion.Focusing on how to improve the accuracy of aviation IR-visible image fusion so as to enhance the reliability of target identification and tracking,the paper studies and achieves high-precision registration for different-source images.To meet the requirements of such fusion,it designs an image fusion method combined with region segmentation that has good target indication and high fusion accuracy.Also,for practical engineering application of the platform,it designs an IR-visible registration method based on optical alignment,and realizes the realtime processing of this fusion system.The main research contents and contributions of this paper are summarized as follows:1.To solve the problems of image distortion in the forms of translation,rotation and zooming caused by UAV vibration,error of sensor focusing and other factors during the image acquisition,this paper designs a high-precision point-feature different-source image registration method based on edge contour description applying to various application scenes by considering the fact that the IR-visible light sensor imaging reflects the spectral characteristics of different wavebands and thus causes a big gray difference between images.First,in the feature extraction phase,use the multiscale corner detection method to detect salient corner points on the edge contour of different-source images,define the principal direction of feature points,and obtain salient feature points containing the information on scale,position and direction.Then in the feature description phase,use the method of feature point description based on shape features to generate the shape descriptors,and zoom out the search window of feature description in order to improve the algorithm efficiency.Finally,use both fast nearest-neighbor search and RANSAC algorithm to rapidly match the feature points and weed out mismatched point pairs in order to improve the registration accuracy.The experimental result shows that,the proposed registration method can effectively improve the accuracy of extracting the features of the same name from different-source images,meet the demand of high-precision IR-visible image registration in various application scenes,and enhance the operation efficiency while ensuring the registration accuracy.2.To solve the problems on IR-visible fusion images such as blurred target background information,less salient target indicators,and susceptibility to noise jamming caused by the complex background and low SNR of aviation IR images,this paper proposes a region segmentation strategy based on significance guidance by considering the fact that the hot-target information of IR images is highlighted.First,conduct the significance extraction of an IR image based on global contrast.Then segment the image in accordance with the significance result.As the hot-target information of the IR image is highlighted and its background information is blurred after the significance extraction,the contrast of the whole image will be enhanced.Therefore,this segmentation strategy can highlight the target region in the IR image and effectively suppress the influence of noise on IR image segmentation.3.To solve the problems the IR-visible fusion images on an airborne photoelectric platform are faced with,such as bad contrast,low fusion accuracy,and susceptibility to noise jamming,this paper proposes an IR-visible image fusion method combined with region segmentation in the DTCWT(dual-tree complex wavelet transform)region.First,segment the region in an IR image by significance to identify the region of interest(ROI)and the background region,and map the segmentation result to the visible light image.Then fuse the low-frequency components in the DTCWT region according to the region segmentation result.For high-frequency components,assign the region weights by the information richness of region details to conduct fusion based on both weights and adaptive phases,and then introduce a shrinkage function to suppress noise.Finally,reconstruct the fused low-frequency and high-frequency components to obtain the fusion image.The experimental result shows that,the proposed fusion method can fully extract complementary information from the source images to obtain a fusion image with good target indication and rich information on scene details.Moreover,this method is endowed with good stability and high fusion accuracy.Therefore,it can meet the fusion requirements of IR-visible image fusion system.4.To meet the real-time fusion demand of IR-visible image fusion system,this paper designs the image fusion hardware platform based on FPGA.According to the requirements of this hardware platform and its practical engineering application,we use optical alignment and the translation marking of zoom system to calculate the registration parameters of different-source images,and then define the fusion algorithm in line with the realtime requirement.Finally,we verify the fusion system experimentally,which proves to be able to satisfy the practical engineering demand by providing a good fusion effect and the required realtime processing.