Skip to content

Anomaly detection in the surveillance videos under varying visibility (day-night and sunny-rainy)

License

Notifications You must be signed in to change notification settings

lyn1874/anomaly_detection_under_varying_visibility

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Anomaly detection under varying illuminations and weather conditions

The goal of this repository is to evaluate the performance of the state-of-the-art anomaly detectors on the scenes with varying visibility. We consider the visibility changes are caused by illumination changes (day-night) and weather switches (sunny-rainy). The state-of-the-art anomaly detectors that we use are listed below:

Prepare the datasets

Most of the anomaly detection benchmark datasets contain relatively clean frames, which are usually collected under a well-controlled environment (sunny and similar times of the data). These frames are not very representative of the surveillance cameras in a city-context where the illuminations and weathers are frequently changing. Thus, to quantitative evaluate the effectiveness of the anomaly detectors under scenes with varying visibility, we augment the standard Avenue dataset by altering illuminations and adding raindrops.

  • Download the dataset

    ./prepare_dataset.sh

  • Augment the dataset

    python3 aug_data.py --rain_type heavy --bright 4 --datapath datadir

Train and evaluate the models

In each subfolder, first run ./requirement.sh to install the required packages. Then run ./run.sh to train and model and ./eval.sh to evaluate the model. Note the datapath argument in each subfolder needs to be manually defined

References

https://github.com/feiyuhuahuo/Anomaly_Prediction

https://github.com/cvlab-yonsei/MNAD

https://github.com/donggong1/memae-anomaly-detection

About

Anomaly detection in the surveillance videos under varying visibility (day-night and sunny-rainy)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published