Enhanced Feature Matching in Image Stitching Using ORB for Image-Based Mapping Using UAV

Abstract Image-based region mapping requires an image-stitching process to produce a complete map while preserving the details of the mapped objects. Image acquisition can be done using UAVs such as drones, and image stitching can be performed using the Oriented FAST and Rotated BRIEF (ORB) techniques. However, this algorithm has a drawback: There are many mismatches during feature matching. In this study, a maximum pixel displacement constraint between images was applied during the feature-matching process. Applying this constraint improved the feature-matching results due to a significant reduction of the mismatch rate from 84.28% to 25.17%. This reduction positively impacted subsequent stages, including map stitching. The mismatch removal process and the acquisition of transformation models during image registration became faster, reducing the processing time from 17m 36.8s to 1m 17.2s. Additionally, there was an improvement in the SSIM metric score, indicating that the stitched image results had a higher similarity to the reference image captured by a digital camera, increasing from 0.8423 to 0.8489.