Research on Panorama Reconstruction Technique of UAV Aerial Image Based on Improved ORB Algorithm

Classic ORB (oriented FAST and rotated BRIEF) algorithm usually extracts feature points globally, which will result in more false feature match pairs and longer computation time. In this paper, an improved ORB algorithm is proposed to reconstruct the panorama of aerial images. First, phase correlation method is used to obtain overlapping area between to-be-stitched image and reference image, and improved ORB is adopted to detect and describe feature points on overlapping regions. Second, K-nearest neighbor (KNN) algorithm is utilized to match feature points roughly, and false matching points are removed by Random Sample Consensus (RANSAC) algorithm. Finally, the improved weighted average fusion algorithm is adopted to eliminate ghosting, while realizing the seamless splicing of UAV images. Experimental results show that, compared with SIFT and classic ORB, the algorithm proposed in this paper can obtain higher registration accuracy, and also significantly improve the stitching accuracy and efficiency of UAV aerial images.