Comparative Studies of Descriptor-Based Image Matching Techniques for AAV Applications

Autonomous aerial vehicles (AAVs) require high operational reliability. One of the key issues is ensuring uninterrupted localization. Unfortunately, the Global Navigation Satellite System (GNSS) is susceptible to interference. This problem can be overcome by redundant localization using vision, where the AAV position is determined by comparing satellite images assigned to a given area with images recorded by an onboard camera. It is a real-time process, so a reliable image-matching technique is crucial. This article presents the results of comparative studies of typical image feature-matching algorithms. Pre-processed images recorded during flights and satellite orthophotomaps are used for these studies. The comparison results are presented graphically. The work consists of three stages: – image labelling, – automatic image-matching, and – analysis of the results. The research goal is to identify the image-matching method that gives the most precise result and is not computationally demanding. The data were collected during flights at various altitudes, with various tilt angles and route lengths. The data diversity allows us to assess the quality of map-matching methods, considering various flight parameters. The flights were carried out in good weather conditions. The most effective image-matching methods, such as DKMv3 and ROMA, were indicated, and their ability to perform efficient matching was highlighted. Additionally, the limitations of vision-based localization, including challenges related to illumination changes, occlusions, and viewpoint variations, are discussed. The presented research is relevant for designing resilient localization systems.