Video Anomaly Detection using Variational Convolutional LSTM Autoencoder

Unintentional, inadvertent, unanticipated, or unplanned events are referred to as anomalies or abnormal events. Anomaly detection in surveiiance video has been a topic of active research for several years and is a very challenging task. In this research, we investigate the performance of the variati...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2023 International Conference on Communication, Circuits, and Systems (IC3S) S. 1 - 5
Hauptverfasser: Sahoo, Sandhya Rani, Kokkiligadda, Jaideep, Dash, Ratnakar
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 26.05.2023
Schlagworte:
and Video anomaly detection
Circuits and systems
Gaussian distribution
Integrated circuit modeling
KL divergence loss
Long short-term memory (LSTM)
Mean square error methods
Spatiotemporal phenomena
Task analysis
Variational autoencoder(VAE)
Video sequences
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Unintentional, inadvertent, unanticipated, or unplanned events are referred to as anomalies or abnormal events. Anomaly detection in surveiiance video has been a topic of active research for several years and is a very challenging task. In this research, we investigate the performance of the variational autoencoder taking into account convolutional LSTM. To improve model performance, the model makes use of a latent space that follows a Gaussian distribution. Instead of just employing mean square error, the loss function used in this study accounts for both mean square error and KL divergence loss. Combining both losses results in a more effective regularisation impact of the model. To show the model's resilience, this study takes into account two publicly accessible datasets: the Avenue dataset and the UCSD Ped1 dataset. Additionally, a performance comparison is made between convolutional LSTM and variational convolutional LSTM autoencoder.
DOI:10.1109/IC3S57698.2023.10169381