An Automated Monitoring System for Abnormal Event Detection
- Resource Type
- Conference
- Authors
- Pervaiz, Mahwish; Akhter, Israr
- Source
- 2022 International Conference on Electrical Engineering and Sustainable Technologies (ICEEST) Electrical Engineering and Sustainable Technologies (ICEEST), 2022 International Conference on. :1-6 Dec, 2022
- Subject
- Bioengineering
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Fields, Waves and Electromagnetics
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Event detection
Surveillance
Sociology
Feature extraction
Wavelet analysis
Rail transportation
Security
abnormal event classification
gray wolf optimizer
region shrinking
xg-Boost classifier
- Language
Public events are source of entertainment for every people. Number of such events have been increased with increase in population. Security is the main concern of people especially in such events. Multiple surveillance systems have been utilized to keep track of security concerns and keep an eye on activities of crowd. These systems help to track people and identify suspicious activities or unexpected events occurs. In this paper, a novel method has been proposed to detect abnormal activity occurred at any in-door/out-door environment. Initially, Gaussian filter has been used as preprocessing to detect the foreground objects and background removal. Then fuzzy c-mean has been opted to verify human silhouettes and shadow removal has been performed. After that novel method for region shrinking has been implemented to isolate occluded humans and feature descriptor comprised of velocity and wavelet analysis has been extracted for each silhouettes. Feature has been optimized using Gray Wolf optimizer and abnormal event classification is performed using the XG-Boost classifier. Performance is evaluated using UMN dataset and UBI-Fight dataset, each having different a nature of anomaly. The mean accuracy for the UMN and UBI-Fight datasets is 90.14%, and 81.3% respectively. These results are more accurate as compared to other existing methods.