Online Rain/Snow Removal from Surveillance Videos (TIP2021)

This paper proposes a new online rain/snow removal method from surveillance videos by fully encoding the dynamic statistics of both rain/snow and background scenes in a video along time into the model, and realizing it with an online mode to make it potentially available to handle constantly coming streaming video sequence.

Minghan Li, Xiangyong Cao, Qian Zhao, Lei Zhang, Deyu Meng

Please go to the Homepage to obtain more information about our work.

Overview

The diagram of the proposed OTMS-CSC model implemented on a video with dynamic background.

Installation

1. Clone this repo

2. Install Matlab

3. Compile GCO-v3.0

• Download gco-v3.0 library, and unzip the file.

• Start Matlab, and make gco-v3.0\matlab your working directory or add it to your path.

• To test your installation of GCO_MATLAB, run the gco-v3.0\matlab\GCO_UnitTest command.

4. Compile spams-matlab

• Please complie it according to spams-matlab/README.md

5. Download dataset (NTURain or your own videos) and put the file into the input folder

6. Run demo.m

YTVOS2019-Rain dataset

• We build YTVOS2019-Rain dataset for video rain removal verification on video instance segmentation task. You can download YTVOS2019-Rain dataset from here.

Citation

Please cite our paper if you find anything helpful,

@article{Li2021OnlineRR,
  title={Online Rain/Snow Removal From Surveillance Videos},
  author={Minghan Li and Xiangyong Cao and Q. Zhao and L. Zhang and Deyu Meng},
  journal={IEEE Transactions on Image Processing},
  year={2021},
  volume={30},
  pages={2029-2044}
}

Contact

For more information please contact [email protected] or [email protected].