A Military Image Stitching Method with Key Targets Constraints Based on Semantic Segmentation

compared with commercial images, military images usually have large parallax and depth of field due to the complex battlefield environment. As a result, key targets in the image, such as fighters, weapons, vehicles, military facilities, etc., tend to occupy a smaller percentage of pixels. Correspondingly, blur or ghosting will occur after image stitching which may lead to misdetection and misjudgment to commanders in a war. Therefore, a novel method for military images with key targets constraints is proposed based on semantic segmentation. First, the whole image is distinguished to several regions according to the kind of the military targets by a semantic segmentation. Second, the homographies of all the local regions of targets are calculated and added to the background homography as constraints to obtain the global homography of the whole image. Finally, a new image is obtained by stitthing the two images transformed by the global homography. Experiments show that the proposed method can improve the articulation of key objects in military images which can provide accurate information for the commanders in the complex battlefield environment.