Feature selection algorithm assisted residual channel attention spatio-temporal auto encoder for video anomaly detection

The common use of video cameras in public and private spaces for traffic management, security, and anomaly detection has brought significant benefits. However, automatically detecting anomalous activity within this vast amount of video data offers a significant challenge. Existing video anomaly dete...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Signal, image and video processing Jg. 19; H. 2; S. 148
Hauptverfasser: Prasudha, M. Lakshmi, Sukhavasi, Vidyullatha, Neha, Kandula, Lunawat, Poonam Shaylesh
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.02.2025
Springer Nature B.V
Schlagworte:
Accuracy
Algorithms
Anomalies
Artificial intelligence
Automation
Brightness
Cameras
Computer Imaging
Computer Science
Crime
Datasets
Deep learning
Feature selection
Frames (data processing)
Image Processing and Computer Vision
Multimedia Information Systems
Neural networks
Optimization
Original Paper
Parking violations
Pattern Recognition and Graphics
Root-mean-square errors
Signal,Image and Speech Processing
Surveillance
Video data
Violence
Vision
Cosine similarity
Keyframes
Autoencoder
Video
Residual channel attention
Chaotic gazelle optimization
Anomaly
ISSN:1863-1703, 1863-1711
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The common use of video cameras in public and private spaces for traffic management, security, and anomaly detection has brought significant benefits. However, automatically detecting anomalous activity within this vast amount of video data offers a significant challenge. Existing video anomaly detection algorithms have limitations such as high computational cost, low performance, and a large demand for specific hardware resources. To overcome these drawbacks, this paper introduces a novel autoencoder model for accurate anomaly detection. First, the University of Central Florida (UCF) crime dataset is used to collect input data. Input videos are transformed into frames; then, keyframes are chosen from the Cosine Similarity distance measure. Selected keyframes are fed into the pre-processing stage to improve the brightness-guided pre-processing procedure for enhancing the brightness variability of keyframes. Optimal features from the pre-processed frames were taken using the chaotic gazelle optimization algorithm (CGOA) to reduce training time. Finally, residual channel attention assisted spatial–temporal autoencoder (RCAA-STAE) in detecting anomaly classes. The performance of the proposed model is estimated and compared with previous studies. The proposed model attained effective performance results in terms of accuracy, precision, recall, specificity, kappa, Root mean square error (RMSE), area under the curve (AUC), and mean square error (MSE). Accuracy performance attained by the proposed approach is 99.23% and 99.5% for normal and abnormal classes using the UCF-crime dataset.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-024-03640-0