Research On Anti-viewpoint Changing Image Registration Algorithms Based On Features

Along with the improvement of the status of precision-guided weapons inmodern warfare, the development of image registration has been paid more attentionby national military-industrial departments. With thorough application of it, imageregistration technology need to be more accuracy and efficiency. Locate the targetquickly and accurately can reserve adequate time for attitude adjustment of the air-to-ground seeker. So carry out the search on anti-viewpoint changing imageregistration algorithms has important significance.The paper focuses on anti-viewpoint changing image registration algorithmsbased on features.Following work has been done:First of all, systematic exposition mainly technical aspects of feature-basedimage registration algorithm: we discuss the camera motion and coordinatetransformation model of the image space, which are closely related to imageregistration firstly; then, we introduce the methods to detect, describe and matchaffine invariant features; related methods such as interpolation is discussed atlast.This part is the foundation of the following work.Secondly, to overcome the great viewpoint changing between the real-timeimage which is acquired by the air-to-ground seeker and the reference image oftarget area during the electro-optical imaging guidance, an improved SIFT algorithmon scene matching is proposed to bridge the gap of the original one, which may leadto failed scene matching for the lack of correct matched pairs. It increases matched pairs by extracting viewpoint changing resistant points which got from the affinetransformation resistant regions extracted in the Gaussian scale-space. It has beenproved that it can acquire nearly5times more correct rate of matching than theoriginal one with the viewpoint ranges from50°to more than60°and can satisfy thepractical need.Thirdly, to overcome the great viewpoint changing between images to beprocessed, an image matching algorithm based on speed up robust feature has beenproposed. The proposed algorithm simulated the process of viewpoint changing ofreference image in order to obtain a group of images, and extracted speed up robustfeatures in them. Finally, we can get the matching point pairs by the method of theeuclidean distance criterion. It has been proved that proposed one can increase thematching pairs and overcome the viewpoint changing in during automatic targetrecognition of the air-to-ground seeker.Only with the help of hardware platform can we meet the requirement ofregistration algorithm of the air-to-ground seeker during its automatic targetrecognition.We also designed the target recognition and tracking system of the seeker,and transplanted the proposed algorithm successfully into it. Experimental resultsshow that the hardware platform can meet the expected performancerequirements.And the SURF-based image registration algorithm resist perspectivetransformation, which performs stably.