An Efficient and Fast Image Mosaic Approach for Highway Panoramic UAV Images

Abstract

Nowadays, real-time monitoring of highway operation by unmanned aerial vehicle (UAV) technology is one of the research frontiers for urban remote sensing. In general, the existing stitching algorithms can meet the basic requirements in terms of accuracy, but their splicing speed cannot meet the real-time stitching requirements of UAV. The cause is that the time consumption sharply increases when stitching plenty of UAV images—this is the bottleneck problem. Herein, we proposed a novel splicing method based on the Superpoint network and a self-designed algorithm of matrix iteration. In this method, we take advantage of an advanced deep learning algorithm—Superpoint to efficiently extract image feature points for calculating the geometric transformation matrix, and make the Superpoint model more suitable for highway. More importantly, for the purpose of further improving the stitching speed and realizing real-time stitching for a large number of UAV images, we specially designed an algorithm of matrix iteration to accurately represent the image transformation relationships, i.e., a matrix is iterated through each adjacent transformation matrix relationship. It is the first time that an algorithm of transformation matrix iteration has been designed to address the bottleneck problem in stitching plenty of UAV images. As a result, the experiments indicate that the proposed method has remarkably enhanced the stitching speed and accuracy for plenty of UAV images. Notably, even in the condition of no air triangulation parameters, it can realize real-time stitching.

Keywords

• Image mosaic
• matrix iteration
• real-time stitching
• unmanned aerial vehicle (UAV)