Feature-Based Stitching Algorithm of Multiple Overlapping Images from Unmanned Aerial Vehicle System

Abstract

This research presents a novel solution to cluster images from an Unmanned Aerial Vehicle System (UAV) using a feature-based stitching algorithm that employs a Binary Robust Invariant Scalable Keypoints (BRISK) feature detection technique. The UAV images are clustered by comparing the degree of overlaps, via the BRISK feature detection and matching, and a set threshold. The clustering serves as an essential solution in solving the stitching errors of UAV images, which are predominantly caused by inconsistent or overlapping regions. The Random Sample Consensus (RANSAC) is able to eliminate outliers on every paired keypoint and produces a precise computed homography or displacement of objects within images, thus making the stitching of multiple overlapping UAV images a success. Since the goal is to detect and cluster images qualified for image stitching. The aggregation of image smoothing is recommended. This will have a significant effect in eliminating minor problems such as the lack of color balance and seam presence, problems that slightly affect the visual appearance of the stitched UAV images.