Comparison of Feature Extraction and Matching Techniques Applied to Monocular UAV Images With Low-quality and Land Cover Variations

  • Nathan A. Z. Xavier UFMG
  • Jonathan De A. Lapa ITA
  • Douglas D. De C. Honório IEAv
  • Rafael C. De Oliveira ITA
  • Elcio H. Shiguemori IEAv
  • Marcos R. O. A. Maximo ITA
  • Marco A. P. Domiciano IEAv

Resumo


Extracting features from images is a recurring task in applications highly dependent on cameras, mainly on UAV visual navigation. By using sequential images, on different land covers and low-quality images, variable performance is expected based on both modification characteristics. This study applies the techniques GFTT, SIFT, MSER, FAST, BRISK, ORB, AKAZE, and SuperGlue on the dataset KDD-BR 2022, which contains multiple UAV images of various land covers and low-quality images. The performance of each technique on feature extracting and matching is presented while verifying the methods that better generalize this process.
Palavras-chave: Feature extraction, Feature matching, Computer vision, KDD, UAVs, Visual odometry
Publicado
09/10/2023
XAVIER, Nathan A. Z.; LAPA, Jonathan De A.; HONÓRIO, Douglas D. De C.; OLIVEIRA, Rafael C. De; SHIGUEMORI, Elcio H.; MAXIMO, Marcos R. O. A.; DOMICIANO, Marco A. P.. Comparison of Feature Extraction and Matching Techniques Applied to Monocular UAV Images With Low-quality and Land Cover Variations. In: SIMPÓSIO BRASILEIRO DE ROBÓTICA E SIMPÓSIO LATINO AMERICANO DE ROBÓTICA (SBR/LARS), 15. , 2023, Salvador/BA. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2023 . p. 47-52.