EVALUATION OF IMAGE MATCHING ALGORITHMS IN PROCESSING UNMANNED AERIAL VEHICLE (UAV) IMAGES

Abstract

Recent years have witnessed the fast developments of unmanned aerial vehicles (UAVs). As an alternative to traditional image acquisition methods, UAVs bridge the gap between terrestrial and airborne photogrammetry and enable flexible acquisition of high-resolution images. Image matching plays an important role in several applications of computer vision. Image matching can be seen as the process of finding conjugate points in two or more overlapping images automatically and this process serves as one of the fundamental tasks in digital photogrammetry. Towards achieving the aim of this research work, the performance of three different image matching algorithms was evaluated and compared. These are Speed Up Robust Features (SURF), Features from Accelerated Segment Test (FAST) and Binary Robust Invariant Scalable Keypoints (BRISK) on images acquired with an Unmanned Aerial Vehicle (UAV) and these images were used to verify the effectiveness of the image matching algorithms and coordinates were also measured and extracted from the images. The data used was acquired from a secondary source. The dataset consists of a pairs of overlapping images which are aerial photograph of Federal University of technology, Akure (FUTA). The left image of stereopair (Image 1) is 1080*108, 101KB and the right image of stereopair (Image 2) is 1080*717, 121KB. The evaluation and comparison of the algorithms were based on the number of points or features detected, number of features matched, total image matching time and efficiency. It was shown that SURF is the most efficient algorithm with 72% efficiency in terms of correctness and speed of execution. From the statistical analysis carried out, the standard deviation of pixel coordinate extracted from images based on SURF algorithm with value of X= 0.1017 and Y= 0.1110 was the least. As part of future work, it is recommended to use more image matching algorithms in addition to the ones used in this research using the same dataset.