Machine Learning-Based Threat Detection in Crowded Environments
Cherin Yacoob Wattacheril, G. R. Hemalakshmi, Ashin Murugan, P Abhiram, G Harisankar, A. Mashour George · 2024
This paper presents the design and implementation of an intelligent system that integrates machine learning and computer vision to create an advanced framework for detecting anomalies and threats in crowded public spaces. The system continuously analyzes video feeds from strategically placed cameras, processing visual data in real-time to identify unusual or suspicious activities. Leveraging advanced machine learning algorithms, the system processes these video streams by learning from historical data and adapting to new patterns over time. It is capable of identifying events such as unattended objects, aggressive behaviour, or unauthorized access. The system builds behavioural profiles of individuals within the monitored environment, analyzing factors such as movement patterns, interactions, and crowd dynamics. Alerts are triggered whenever deviations from expected behaviors are detected. The system also prioritizes privacy by anonymizing individuals, focusing on tracking movement patterns rather than facial recognition. It ensures compliance with privacy regulations through robust safeguards. Upon detecting an anomaly, the system either notifies security personnel or activates automated response mechanisms, facilitating rapid interventions such as deploying security teams or securing specific areas. The design is scalable, allowing for coverage of large public spaces like airports, stadiums, and shopping centers, and can adapt to varying environmental conditions and crowd densities. Continuous learning from false positives and negatives enhances algorithm performance, ensuring ongoing improvement. Regular audits and updates are also conducted to maintain optimal functionality.