OSMO: Online Specific Models for Occlusion in Multiple Object Tracking under Surveillance Scene

Created by Xu Gao, Peking University

Introduction

OSMO is an online multi-object tracking framework which deals with two kinds of occlusion in the surveilalnce scene.

Citation

If you find OSMO useful in your research, please consider citing:

 @inproceedings{osmo2018,
    author = {Gao, Xu and Jiang, Tingting},
    title = {OSMO: Online Specific Models for Occlusion in Multiple Object Tracking Under Surveillance Scene},
    booktitle = {Proceedings of the 26th ACM International Conference on Multimedia},
    year = {2018},
    pages = {201--210},
} 

Usage of the demo

1. Our project is written in Python 2.7 and PyTorch 0.2. Please set the environment at first.

2. Download the CampusStone dataset from https://drive.google.com/open?id=1nL60VdWkOkvjkdkAY53cuuLbV6wBStSn and unzip the file.

3. Download the code using "git clone https://github.com/gaoxu1024/OSMO.git".

4. Put the unziped dataset into ./dataset/.

5. For testing, use "sh osmo_run_list.sh".

We provide our tracking results in the folder ./result/. Besides, we only present the CampusStone dataset here currently, since other three datasets may have some private problems with the government.

References

[1] Xu Gao, and Tingting Jiang. OSMO: Online Specific Models for Occlusion in Multiple Object Tracking Under Surveillance Scene. Proceedings of the 26th ACM International Conference on Multimedia (MM), 2018.

Contact

If you find any bug or issue of the software, please contact gaoxu1024 at pku dot edu dot cn