Multi-Date Earth Observation NeRF: The Detail Is in the Shadows

Roger MarĂ­, Gabriele Facciolo, Thibaud Ehret; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 2035-2045

Abstract


We introduce Earth Observation NeRF (EO-NeRF), a new method for digital surface modeling and novel view synthesis from collections of multi-date remote sensing images. In contrast to previous variants of NeRF proposed in the literature for satellite images, EO-NeRF outperforms the altitude accuracy of advanced pipelines for 3D reconstruction from multiple satellite images, including classic and learned stereovision methods. This is largely due to a rendering of building shadows that is strictly consistent with the scene geometry and independent from other transient phenomena. In addition to that, a number of strategies are also proposed with the aim to exploit raw satellite images. We add model parameters to circumvent usual pre-processing steps, such as the relative radiometric normalization of the input images and the bundle adjustment for refining the camera models. We evaluate our method on different areas of interest using sets of 10-20 pre-processed and raw pansharpened WorldView-3 images.

Related Material


[pdf]
[bibtex]
@InProceedings{Mari_2023_CVPR, author = {Mar{\'\i}, Roger and Facciolo, Gabriele and Ehret, Thibaud}, title = {Multi-Date Earth Observation NeRF: The Detail Is in the Shadows}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {2035-2045} }