Big Data-Driven Video Anomaly Detection Using VideoMAE for Visual Analytics in CCTV Surveillance

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Traditional manual surveillance is inefficient and prone to human error, especially in high-stakes environments.
• The increasing need for robust security in sensitive organizations necessitates advanced monitoring solutions.
• Current security systems struggle to cope with the volume of video data, risking missed critical events.

Purpose of the Study:

• To develop an intelligent artificial intelligence (AI)-powered system for automated detection of anomalous activities in surveillance footage.
• To enhance the efficiency and accuracy of security monitoring in confidential and highly sensitive organizations.
• To provide a proactive and seamless monitoring solution that overcomes human limitations in traditional surveillance.

Main Methods:

• Utilized a deep learning framework, specifically the Video Masked Autoencoder model, for video understanding.
• Employed transformer models, including Video Vision Transformers, for effortless detection of suspicious activity.
• Developed a user-friendly web interface for uploading footage and receiving anomaly detection results.

Main Results:

• The AI-driven system successfully spots anomalous activities in surveillance footage without manual intervention.
• The system accurately identifies the type of anomaly and the specific time frame of its occurrence.
• Demonstrated a significant improvement in the speed and efficiency of detecting suspicious activities compared to manual monitoring.

Conclusions:

• The developed AI system offers a more effective and efficient solution for surveillance in high-stakes environments.
• Automated anomaly detection using deep learning and transformer models significantly enhances security.
• The system provides proactive, accurate, and stress-free monitoring, reducing the risk of critical events being missed.