The Performance Evaluation of Image Matching Techniques Within Uav Images

Abstract

It was aimed that the images were acquired with two different types of UAV by featurebased transformation algorithms such as SURF (Speeded Up Robust Features), FAST (Features from Accelerated Segment Test) and BRISK (Binary Robust Invariant Scalable Keypoints) in this study. Images (acquired by both UAV types) grouped by inclination. This classification is based on the wing type of the UAV (Rotary-Wing UAV” and “FixedWing UAV). Images with different characteristics were used to produce mosaics from the algorithms. The first performance preferred a flight height of 30 m (Ground Sample Distance, 0.82 cm/pixel) with the frontal overlap of 80%, and the second performance preferred a flight height of 60 m (GSD, 1.64 cm/pixel) and same overlap. Ten images from both performances were combined in all algorithms. Mismatches have been observed, and the mosaics produced after a very long process are not found satisfactory. According to the results, for rotary-wing UAV (SURF, BRISK and FAST), the algorithm run times were determined as 76.5 minutes, 11 minutes and 1839 minutes. Also, for fixed-wing UAV (SURF, BRISK and FAST), algorithm run times of 238 minutes, 95 minutes and 3350 minutes were determined.