Abnormal Activity Recognition from Surveillance Videos Using Convolutional Neural Network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Millions of pilgrims attend the Hajj annually, necessitating robust crowd monitoring for safety.
• Existing surveillance systems rely on manual monitoring, which is resource-intensive and inefficient for large-scale events.
• Closed-circuit television (CCTV) cameras generate vast amounts of data, posing challenges for real-time analysis.

Purpose of the Study:

• To develop an intelligent and automatic system for efficient crowd monitoring and abnormal activity detection.
• To address the limitations of existing methods in extracting discriminative features from surveillance videos.
• To enhance the safety and security of pilgrims during large gatherings.

Main Methods:

• A lightweight Convolutional Neural Network (CNN) model was trained on a pilgrim dataset for initial pilgrim detection.
• Spatial features were extracted using the lightweight CNN model.
• A Long Short-Term Memory (LSTM) network was employed for temporal feature extraction.
• The integrated system generates real-time alarms upon detecting violent activity or accidents.

Main Results:

• The proposed model achieved 81.05% accuracy on the Surveillance Fight dataset.
• The model demonstrated 98.00% accuracy on the Hockey Fight dataset.
• The system effectively identifies and alerts authorities to abnormal activities in real-time.

Conclusions:

• The developed lightweight CNN-LSTM framework offers an accurate and efficient solution for abnormal activity detection in surveillance videos.
• This system can significantly aid law enforcement in preventing accidents and stampedes.
• The approach provides a scalable and effective method for crowd management in large-scale events.