Research On The Registration And Mosaic For UAV Infrared Image Sequence

Combined with the advantages of UAV and infrared detection technology,such as UAV’ssmall size,light weight,responsive,well concealed,inexpensive and infrared detection technolog-y’s long detection distance,good penetrating ability,strong anti-interference ability, all-timework.etc,the technology based on UAV to obtain infrared images has been widely used inenvironmental monitoring,resource exploration,military reconnaissance and emergency support,but the obtained infrared image sequence with these characteristics such as small field ofview,serious noise pollution,image blur,rotation and scaling of geometric distortion has seriouslyrestricted the scope of application. In order to eliminate the negative factors in the image and getan ideal, reliable large field infrared scene image for subsequent processing, this paper hasstudied the registration and mosaic technology for the UAV infrared image sequence, the mainlydone work and obtained results just as follows:1.Introduced the background,significance,domestic and international research in the field ofinfrared sequence image registration and mosaic.2.Studied the denoising technology for infrared image sequence.Respectively introduced theaccess mechanism and the characteristics of the infrared image,the noise types and distribit-ioncharacteristics,wavelet transform theory,Bayesian and denoising quality assessment.Based on thetheory of Bayesian wavelet denoising,this paper used redundant wavelet transform and bivariatejoint shrinkage function to enhance the wavelet coefficients correlation in the sub-band andsubband to achieve local adaptive threshold denoising. Experiments show that this method hasbetter effect than traditional algorithms..3.Combined with the characteristics of the infrared image sequence,this paper has studiedtwo scale-invariant feature extraction operators-SIFT and CenSurE operators,image matchingcriteria,search algorithms,mismatch removed.etc.Combining CenSurE， linear search andRANSAC operators together as the method to complete feature extraction and matching work.4.Studied the geometry registration for the infrared image sequence, based on the roughregistration,this paper has introduced the KLT tracking theory to solve the problem of finealignment between the images,and then used sparse bundle adjustment as the global alignmentstrategy to eliminate the dislocation, distortion and ghost in the infrared image sequence mosaicprocess. Experiments show that this method can effectively accomplish the mosaic work of theinfrared image sequence.