This repository contains the code and scripts for the paper Is it Raining Outside? Detection of Rainfall using General-Purpose Surveillance Cameras
The paper investigates rain detection with general-purpose surveillance cameras. The previous state-of-the-art method is compared with a 3D Convolutional Neural Network, on data from two different traffic crossings recorded in Aalborg, Denmark.
As this is a collection of research code, one might find some occasional rough edges. We have tried to clean up the code to a decent level but if you encounter a bug or a regular mistake, please report it in our issue tracker.
The evaluation code is built around the AAU VIRADA dataset which is published at Zenodo.
The NVidia Video Loader (NVVL) framework was used for loading video snippets efficiently. The Video Super-Resolution example was used as the basis for the provided docker file
The pytorch implementation of the basis C3D network was derived from David Abati's implementation.
The Bossu implementation utilizes OpenCV v. 3.1.0
All code is licensed under the MIT license, except for the modified NVVL code, which is subject to its own license.
Please cite the following paper if you use our code or dataset:
@InProceedings{Haurum_2019_CVPR_Workshops,
author = {Haurum, Joakim Bruslund and Bahnsen, Chris H. and Moeslund, Thomas B.},
title = {Is it Raining Outside? Detection of Rainfall using General-Purpose Surveillance Cameras},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}