DeepTerRa: Deep Digital Terrain Models from ALS Point Clouds via Rasterization

Abstract

Despite the popularity of deep neural networks in various domains, the extraction of digital terrain models (DTMs) from airborne laser scanning (ALS) point clouds is still challenging. This might be due to the lack of the dedicated large-scale annotated dataset and the data-structure discrepancy between point clouds and DTMs. To promote data-driven DTM extraction, this article collects from open sources a large-scale dataset of ALS point clouds and corresponding DTMs with various urban, forested, and mountainous scenes. A baseline method is proposed as the first attempt to train a deep neural network to extract DTMs directly from ALS point clouds via rasterization techniques, coined DeepTerRa. Extensive studies with well-established methods are performed to benchmark the dataset and analyze the challenges in learning to extract DTM from point clouds. The experimental results show the interest of the agnostic data-driven approach, with submetric error level compared to methods designed for DTM extraction.

Paper

arXiv - IEEE OpenAccess

Data

Download the dataset DeepTerRa v0.1

There are 2 zip files for downloading in bulk or downloading each scene by going into each folder.

The data are stored in *.mat file, which can be read in Python using the scipy package. All the rasterized data are stored as np.ndarray in the read dictionary.

from scipy.io as sio
mat = sio.loadmat('filename.mat')
print (mat.keys())

We have performed the rasterization and offset correction following the discovery mentioned in the paper, so the data sizes are not always round numbers (There are shapes of, eg. 1996x1994). We have made sure all rasters of the same scene stay the same shapes.

Please also let us know by sending email to hoang-an.le[at]irisa.fr if you encounter some problems with the data.

The original las point clouds can be downloaded from the same link, in las_tiles directory. As we have observed an aligning problem when trying to rasterize these point clouds with the original DTMs, and the provided DTMs on our website have been offset to best match with the rasters (see the paper for more information).

Currently known problems

1. DTM of DALES/5145_54340 has a corrupted strip of empty column (zero values)

Source code

The exact code is not not ready to be shared at this moment. However, it is mostly borrowed from the pix2pix code which you can get from the original repo. You’ll need to work on the dataloader to get the rasterizations and DTM from the mat files.

Citation

If you find the material useful please consider citing our work

@article{le22deepterra,
    author = "L{\^{e}}, Ho{\`{a}}ng{-}{\^{A}}n and Guiotte, Florent and Pham, Minh-Tan and Lef{\`{e}}vre, S{\'{e}}bastien and Corpetti, Thomas",
    title={Learning Digital Terrain Models From Point Clouds: ALS2DTM Dataset and Rasterization-Based GAN},
    journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS)},
  year={2022},
  volume={15},
  number={},
  pages={4980-4989},
}