Research on Aerial Image Stitching Technology

Abstract

The rise of UAV has made UAV play a major role in various industries, especially in place of manpower to solve difficult problems such as complex terrain detection and on-site monitoring in disaster areas. The pictures taken by the UAV are spliced to realize the global display of the scene, which is convenient for analysis and research. The integrity and clarity of image stitching depends on the performance of the stitching algorithm. The key to image registration and stitching is the extraction and matching of feature points. The article compares and analyzes the traditional three feature point (key-point) extraction algorithms. The article summarizes their advantages and disadvantages and scope of application. In this paper, SIFT, SURF and ORB algorithms were used respectively to extract feature points from the two images. Then, the nearest neighbor matching method was used to select the optimal matching points to remove the pseudo-matching points and improve the matching accuracy. After image registration, the composite image is prone to splicing gaps and brightness differences due to error accumulation, color differences, and the like. Therefore, in order to make the final panoramic image better, it is more necessary to perform image fusion after image registration processing, correct the difference, and eliminate the stitching gap. In this paper, the improved ORB algorithm combined with the weighted average fusion algorithm is used to achieve smooth transition of the two images. The improved algorithm time is reduced and the efficiency is significantly improved. The experimental results also show that the weighted average algorithm has high effectiveness and practicability in image fusion.