← Back to list of papers of the 2022 EuroGNC conference

CEAS EuroGNC 2022

Spectral Loss for Monocular Self-Supervised Depth and Visual Odometry in Rover Navigation
Juan Ignacio Bravo Pérez-Villar PhD Student, Deimos Space S.L.U., Flight Systems, 28760, Madrid, Spain.
Álvaro García-Martín Associate Professor, Universidad Autónoma de Madrid, Video Processing and Understanding Lab, 28049, Madrid, Spain.
Jesús Bescós Associate Professor, Universidad Autónoma de Madrid, Video Processing and Understanding Lab, 28049, Madrid, Spain.
Abstract:
This article explores the use of monocular self-supervised visual odometry and depth estimation algorithms in rover-like scenarios. The aim of these methods is to learn, using Convolutional Neural Network (CNN) architectures, the estimation of pixel-dense depth maps and visual odometry measurements from monocular video sequences without any associated ground-truth data. The article reviews the core method, its limitations, and the solutions proposed in the literature. Different learning objectives from the literature are tested and the associated results are reported. In addition, a new learning objective based on the frequency domain of the image is proposed to exploit the particularities of the rover-like scenarios, increasing the accuracy of the odometry results.
Keywords: Self-Supervised; Visual Odometry; Monocular; Depth; DCT
 View PDF CEAS-GNC-2022-015


Juan Ignacio Bravo Pérez-Villar, Álvaro García-Martín, Jesús Bescós: Spectral Loss for Monocular Self-Supervised Depth and Visual Odometry in Rover Navigation. Proceedings of the 2022 CEAS EuroGNC conference. Berlin, Germany. May 2022. CEAS-GNC-2022-015.
BibTeX entry:

@Incollection{CEAS-GNC-2022-015,
    authors = {Ignacio Bravo Pérez-Villar, Juan and García-Martín, \'Alvaro and Bescós, Jesús},
    title = {Spectral Loss for Monocular Self-Supervised Depth and Visual Odometry in Rover Navigation},
    booktitle = {Proceedings of the 2022 {CEAS EuroGNC} conference},
    address = {Berlin, Germany},
    month = may,
    year = {2022},
    note = {CEAS-GNC-2022-015}
}