Speeding Up SIFT for Faster Feature-Based UAV Visual Localization

The process of finding and locating an unmanned aerial vehicle (UAV) on a map image in a Global Navigation Satellite System (GNSS)-denied environment is known as vision-based localization. Scale-invariant feature transform (SIFT) algorithm is frequently used for feature extraction and image matching in computer vision. Due to its speed compared to deep learning methods and stability compared to other local feature extraction algorithms like SURF, ORB, and others, the SIFT algorithm is particularly well suited for vision-based localization problems with feature points matching approach. Despite its popularity, SIFT has some processing speed and accuracy issues when used in real-world applications. We use the following methods to improve processing time by speeding up the matching phase: limiting the search area when given latitude and longitude, using the Principal Component Analysis (PCA) algorithm to reduce the dimensionality of the vector descriptor, and performing map image pre-processing to reduce the number of keypoints. The outcomes of this proposed method offer promising improvements in terms of processing time, whilst maintaining a satisfactory level of accuracy.