Research On Infrared Imaging Registration And Stitching Algorithm In Vehicular Moving Platform

Infrared images are widely used in the military field due to their passive work,good antiinterference,strong target recognition and all-weather characteristics.As a popular branch in infrared imaging technology,infrared mosaic technology is an important means to solve the narrow field of view.In the investigation of the military external environment,in order to obtain a large-field operational strategy map,the infrared camera needs to perform allround shooting.Then,the infrared mosaic technology uses to perform high-quality fusion of the images of various angles of view,and the mosaic diagram is presented on the combat podium,and the commander makes accurate and correct decisions according to the displayed information.At present,infrared mosaic technology brings a lot of information convenience to the military battlefield,and it has great application value in the military investigation of the battlefield.Infrared image mosaic technology mainly includes preprocessing of infrared image,feature extraction of image,feature registration of image and image fusion.The main work of present paper is as follow:(1)In the image preprocessing stage,the infrared imaging process is affected by external noise.In order to obtain a complete and accurate mosaic effect picture,the image needs to be denoised.Through the comparison of the effects of three kinds of filtering experiments,this paper uses median filtering to eliminate noise;(2)In the feature extraction stage,this paper mainly studies SIFT features,SURF features and ORB features.It is found that SIFT features have many good properties,but the computational complexity is relatively high.The SURF algorithm is based on SIFT for interest points and feature vectors.Improvements have been made to significantly reduce the speed of calculation;the ORB features have a more pronounced speed advantage than SURF in addition to rotational invariance and scale invariance.In environmental investigation,time is the key to success or failure.Based on real-time,this paper uses ORB features to detect and describe features.(3)In image registration,this paper proposes a bilateral matching algorithm to achieve high-quality matching of features.By comparing the matching performance between RANSAC algorithm and GMS algorithm,this paper uses grid motion-based GMS algorithm for feature matching,and in GMS.The one-way matching principle has been improved,and experimental data shows that the matching accuracy increases by 8%.(4)In the image fusion,for the larger image with overlapping regions,based on the gradual blending algorithm,the problem of blurring around the stitching seam in the previous fusion process is eliminated by revising the overlapping region.And achieved a good stitching effect.