We propose a framework for the detection of 3D objects in point clouds based on a deep learning approach. One of the main issues to achieve this goal is the lack of railway datasets with annotated data. For this reason, the first step of our work is the design and creation of the dataset SARD. This enables the collection of a sufficient number of annotated point clouds to be used in the supervised training of a neural network.
It is composed by: a virtual railway environment generated through the 3D graphics engine Unreal Engine 4, 3D landmarks modelled by using Blender, and the simulation of a LiDAR sensor using the Automated Driving Toolbox, provided by Matlab.
SARD answers to the need of publicly available datasets for the railway scenario, due to their current scarcity. We generated 7500 point clouds and we split them in 50% training and 50% validation sets. We make it freely available to the scientific community at this repository.
You can directly download it from MEGA.nz.
If you use this resource, please cite the following paper:
M. Neri and F. Battisti, “3D Object Detection on Synthetic Point Clouds for Railway Applications,” 2022 10th European Workshop on Visual Information Processing (EUVIP), 2022, pp. 1-6, doi: 10.1109/EUVIP53989.2022.9922901.
BibTex:
@INPROCEEDINGS{Neri_2022_EUVIP,
author={Neri, M. and Battisti, F.},
booktitle={2022 10th European Workshop on Visual Information Processing (EUVIP)},
title={{3D Object Detection on Synthetic Point Clouds for Railway Applications}},
year={2022},
volume={},
number={},
pages={1-6},
doi={10.1109/EUVIP53989.2022.9922901}}
If you have any questions regarding the dataset, you can contact via mail michael.neri@uniroma3.it.