Skip to content

ispc-lab/NeuroAED

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

72 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NeuroAED: Towards Efficient Abnormal Event Detection in Visual Surveillance with Neuromorphic Vision Sensor

This is the NeuroAED dataset and implementation code for the following paper. If you use any of them, please cite:

@ARTICLE{9195516,  
author={G. {Chen} and P. {Liu} and Z. {Liu} and H. {Tang} and L. {Hong} and J. {Dong} and J. {Conradt} and A. {Knoll}},  
journal={IEEE Transactions on Information Forensics and Security},   
title={NeuroAED: Towards Efficient Abnormal Event Detection in Visual Surveillance With Neuromorphic Vision Sensor},   
year={2021},  
volume={16},  
number={},  
pages={923-936},}

NeuroAED dataset

Considering the lack of a neuromorphic benchmark for the abnormal event detection, we record the first neuromorphic vision based abnormal event detection dataset and make it available to the research community at the link: https://pan.baidu.com/s/1eP2DTTzRjlWsZ0xGMyuB0A and the code: 3gpm.

The NeuroAED dataset comprises 152 samples of four different indoor and outdoor scenarios, and is split into four sub-dataset: Walking, Campus, Square and Stair dataset. And each dataset contains two slice sequences: training samples and testing samples. The training samples only contain normal events, while testing samples are both normal and abnormal events. For each slice sample of the NeuroAED dataset, the groundtruth annotation of a binary flag indicating normal or abnormal events occur is provided. With the exception of the Square dataset, the manually generated pixel-level binary masks are contained in each slice sample, which identify the abnormal events regions.

Framework of NeuroAED

We extract the optical flow information from training sample and select activated event cuboids based on the optical flow and event density to locate foreground. For each activated event cuboid, the proposed event-based multiscale spatio-temporal (EMST) descriptor is extracted and feed into models to learn the normal patterns. The trained models are used to identify descriptors of abnormal patterns extracted from the testing sample.

Code Implementation

Preprocess:

a. Extract events information from .aedat file using aedat2txt.py. 
b. Extract optical flow information using jAER.
c. Fix original optical flow file using fix_of.py.
d. The main model file: /Models/core.py.

Running:

a. Generate EMST features using feature_generation.cpp
b. Get the Slice_level and Pixel_level values using /Models/core.py

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published