Real-time aerial image mosaicing using hashing-based matching

Abstract

This paper presents a real-time approach for aerial image mosaicing. On average, a performance of 35ms is achieved in a frame-to-frame process where visual features are extracted in between two images, organised in a database model, sought and matched. For this process to be efficient, we employed binary descriptors, namely ORB descriptors, for rapid feature extraction and fast matching based on the hamming distance, and hash tables as descriptor organisation model where keys for hashing are extracted directly from bit segments of the binary descriptor. By combining these two strategies, we managed to speed up feature matching by 8 times compared to the standard binary descriptor-based feature matching using FLANN in OpenCV. As a result we obtain a rapid image mosaicing, which enables the exploration of the area to be mapped at higher speed of the aerial vehicle.