监控视频异常检测:综述 跟随学习

  • 2021-05-21
  • 机器学习算法分析
  • 公开
简介深度学习方法: 点模型: 聚类判别: 自组织映射(self-organizing map, SOM) Online Learning with Self-Organizing Maps for Anomaly Detection in Crowd Scenes https://ieeexplore.ieee.org/document/5597400 【机器学习笔记】自组织映射网络(SOM) https://www.cnblogs.com/sur

深度学习方法:

点模型:

聚类判别:

自组织映射(self-organizing map, SOM)

Online Learning with Self-Organizing Maps for Anomaly Detection in Crowd Scenes https://ieeexplore.ieee.org/document/5597400

【机器学习笔记】自组织映射网络(SOM) https://www.cnblogs.com/surfzjy/p/7944454.html


生长的神经气(growing neural gas, GNG)

Online growing neural gas for anomaly detection in changing surveillance scenes file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Online%20growing%20neural%20gas%20for%20anomaly%20detection%20in%20changing%20surve.pdf

生长型神经气它可以学习拓扑型数据,是一种具有很好鲁棒性的聚类分析方法。它克服了在绝大多数聚类中需要预先知道聚类数目或者聚类半径的问题。该文提出了一种通过学习数据的拓扑结构解决大数据集分类问题。

有研究将GNG和SVM结合,构建 GNG-SVM 框架。



Gauss混合全卷积VAE(Gaussian mixture fully convolutional VAE, GMFC-VAE)

Video Anomaly Detection and Localization via Gaussian Mixture Fully Convolutional Variational Autoencoder file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Video%20anomaly%20detection%20and%20localization%20via%20Gaussian%20mixture%20fully%20convolutional%20variational%20autoencoder.pdf

无监督异常检测之卷积AE和卷积VAE https://www.cnblogs.com/nanhaijindiao/p/11566725.html

VAE(Variational Autoencoder)的原理 https://www.cnblogs.com/huangshiyu13/p/6209016.html



一类神经网络(one-class neural network, OC-NN)

Anomaly detection using one-class neural networks file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Anomaly%20detection%20using%20one-class%20neural%20networks.pdf

有源码 https://github.com/raghavchalapathy 哈哈哈哈 --- 针对图像的实验



重构判别:

AE:

Learning Temporal Regularity in Video Sequences https://ieeexplore.ieee.org/document/7780455

包含源码 数据 和数据集链接。

卷积自编码器 https://blog.csdn.net/qq_27871973/article/details/90021832 https://blog.csdn.net/qq_33415776/article/details/82668813


Video anomaly detection and localisation based on the sparsity and reconstruction error of auto-encoder https://ieeexplore.ieee.org/document/7492368

通过自动编码器将正常样本表示为稀疏的特征向量。



Limiting the reconstruction capability of generative neural network using negative learning https://ieeexplore.ieee.org/document/8168155

Robust,deep and inductive anomaly detection https://www.semanticscholar.org/paper/Robust%2C-Deep-and-Inductive-Anomaly-Detection-Chalapathy-Menon/1d4ec24a6da3be62dc5d7efbae2a101c63f187e8

DeepFall:Non-invasive fall detection with deep spatio-temporal convolutional autoencoders https://www.semanticscholar.org/paper/DeepFall-Non-invasive-Fall-Detection-with-Deep-Nogas-Khan/1de7d838cfce1c3bef65b134bae3d00b59eddd71

Spatio-temporal autoencoder for video anomaly detection https://www.semanticscholar.org/paper/Spatio-Temporal-AutoEncoder-for-Video-Anomaly-Zhao-Deng/fef6f1e04fa64f2f26ac9f01cd143dd19e549790


VAE:

Generative neural networks for anomaly detection in crowded scenes https://ieeexplore.ieee.org/document/8513816

同时对全局异常和局部异常检测

有github源码


U-Net:

Abnormal event detection in video using generative adversarial nets https://ieeexplore.ieee.org/document/8296547

文章讨论的gan,没有法相u-net相关信息


Training adversarial discriminators for cross-channel abnormal event detection in crowds https://ieeexplore.ieee.org/document/8658774


Robust anomaly detection in videos using multilevel representations file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Robust%20Anomaly%20Detection%20in%20Videos%20Using%20Multilevel%20Representations.pdf

含github源码


生成对抗网络(generative adversarial network, GAN)

Adversarially Learned One-Class Classifier for Novelty Detection https://ieeexplore.ieee.org/document/8578454

源码 https://github.com/khalooei/ALOCC-CVPR2018

针对图像和视频的通用框架


AVID:Adversarial visual irregularity detection file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/AVID-%20Adversarial%20Visual%20Irregularity%20Detection.pdf

ss中提高的源码 https://github.com/masoudpz/AVID-Adversarial-Visual-Irregularity-Detection

异常的检测和定位


Efficient GAN-based anomaly detection file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Efficient%20GAN-based%20anomaly%20detection.pdf

针对图像和网络入侵数据集用来提高检测速度(提高几百倍?)

github源码


GANomaly:Semi-supervised anomaly detection via adversarial training file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/GANomaly-%20Semi-supervised%20anomaly%20detection%20via%20adversarial%20training.pdf

Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Unsupervised%20Anomaly%20Detection%20with%20Generative%20Adversarial%20Networks%20to%20Guide%20Marker%20Discovery.pdf


受限Boltzmann机(restricted Boltzmann machine, RBM)

Energy-based localized anomaly detection in video surveillance https://link.springer.com/chapter/10.1007%2F978-3-319-57454-7_50

可以在流媒体场景中进行增量学习。

受限玻尔兹曼机 https://blog.csdn.net/weixin_42398658/article/details/84279293


联合判别:

同VAE Generative neural networks for anomaly detection in crowded scenes



序列模型:

Abnormal event detection using recurrent neural network https://ieeexplore.ieee.org/document/7810868

逐帧处理并分块。通过预测误差识别异常


A Revisit of Sparse Coding Based Anomaly Detection in Stacked RNN Framework https://ieeexplore.ieee.org/document/8237307

github源码

稀疏编码,堆叠递归神经网络。

提供了一个多视角多监控的数据集。


Decomposing motion and content for natural video sequence prediction file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Decomposing%20motion%20and%20content%20for%20natural%20video%20sequence%20prediction.pdf

STAN:Spatio-temporal adversarial networks for abnormal event detection https://ieeexplore.ieee.org/document/8462388


复合模型:

Autoencoder with recurrent neural networks for video forgery detection file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Autoencoder%20with%20recurrent%20neural%20networks%20for%20video%20forgery.pdf

针对视频拼接伪造的检测。


Abnormal event detection in videos using spatiotemporal autoencoder file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Abnormal%20event%20detection%20in%20videos%20using%20spatiotemporal%20autoencoder.pdf

Anomaly detection in video using predictive convolutional long short-term memory networks file:///E:/--%E8%AE%BA%E6%96%87%E7%9B%B8%E5%85%B3--/%E8%AE%BA%E6%96%87en/Anomaly%20detection%20in%20video%20using%20predictive%20convolutional%20long%20short-term%20memory%20networks.pdf

Remembering history with convolutional LSTM for anomaly detection https://ieeexplore.ieee.org/document/8019325/

卷积LSTM 结合自动编码器。