Título: Landing Aircraft Attitude Estimation From Vanishing Point and Horizon Line Regression
Autores: João Paulo Lara, Gabriel Lott, João Pedro Klock Ferreira, Júlia Santos Moura & Cristiano Leite de Castro
Resumo: Vanishing points and horizon lines carry valuable perspective information about a scene. When the vanishing point location, horizon line slope and camera intrinsic matrix are known, it is possible to infer the camera pose, making these features play important roles in applications such as 3D reconstruction and autonomous navigation. Recent works have employed deep Convolutional Neural Networks (CNN) to estimate these parameters with good performance, but most are limited to ground-level camera scenarios. In this work, we propose an approach to regress the vanishing point coordinates of a runway from frontal images captured by an airplane during the landing phase, as well as to estimate the horizon line slope, using a deep CNN. Furthermore, we leverage vanishing point perspective geometry to infer three degrees of freedom of the airplane—the yaw, pitch and roll angles. The results demonstrate that our approach achieves accurate regression of both the vanishing point and horizon line slope with fast inference times. Moreover, the estimated airplane angles, despite some outliers, show promise for applications in airplane orientation within Autonomous Landing Systems.
Palavras-chave: vanishing point; horizon line; Convolutional Neural Networks; regression; camera orientation; projective geometry.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1191490
Artigo em PDF: CBIC_2025_paper1191490.pdf
Arquivo BibTeX:
CBIC_2025_1191490.bib