Table 1.

Summary of deep learning-based techniques for anomaly detection.

Method	Type	Year	Reference
Convolutional Long Short-Term Memory	RB + Future Frame Prediction	2016	[4]
2D Convolutional Autoencoder	RB	2016	[81]
Sparse Autoencoder	Reconstruction based	2016	[82]
Slow Feature Analysis + Deep Neural Network	Scoring	2016	[9]
Sparse Denoise Autoencoder	Multiclass Classification	2017	[83]
Autoencoder + Cascade Deep CNN	Multiclass Classification	2017	[3]
Spatiotemporal Autoencoder	RB + Future Frame Prediction	2017	[84]
Pretrained DNN + Gaussian classifier	Multiclass Classification	2018	[85]
Autoencoder + Low level features	Reconstruction based	2018	[86]
Multiple Instance Learning	Scoring	2018	[2]
Low-level Features + Autoencoder	Reconstruction based	2018	[86]
Frame predict using GANs	Future Frame Prediction	2018	[43]
Combination of traditional and deep features	Scoring	2019	[87]
Localization feature extraction	Scoring	2019	[88]
AnomalyNet	Reconstruction based	2019	[89]
Optical Flow + Multiple Instance Learning	Scoring	2019	[90]
Social Force Maps + Multiple Instance Learning	Scoring	2019	[91]
Attention module + Autoencoder	Reconstruction based	2019	[92]
Component Analysis + Transfer Learning	Multiclass Classification	2019	[93]
Object detection using SSD + Autoencoder	Multiclass Classification	2019	[94]
Sparse coding Deep neural network	Scoring	2019	[95]
Adaptive Intra-Frame Classification Network	Classification	2019	[96]
Autoencoder + Gaussian Mixture Model	Scoring	2020	[97]