International Research Journal of Engineering and Technology (IRJET) Volume: 12 Issue: 05 | May 2025
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
DEEP LEARNING-DRIVEN SURVEILLANCE SYSTEM FOR ANOMALY DETECTION IN CROWDED ENVIRONMENTS Rutuja Dhumal1, Prof. P. N. Kadam 2, Payal Chandgude3, Sakshi Jamdade4, Maithili Pise5 1,3,4,5 Student, Department of Computer Engineering, SVPM’s College of Engineering, Malegaon BK, Maharashtra,
India.
2Assistant Professor, Department of Computer Engineering, SVPM’S College of Engineering Malegaon BK,
Baramati, Maharashtra, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - With increasing urbanization and crowd density,
in videos. Our system utilizes Convolutional Neural Networks (CNNs) trained to detect unusual body poses or movements from real-time CCTV footage, and immediately notifies the concerned administrator upon detection.
ensuring public safety has become a critical concern. This project proposes an intelligent, deep learning-based surveillance system for real-time detection of suspicious activities in crowded environments such as malls, bus stations, and airports. The system uses convolutional neural networks (CNNs) to process CCTV footage and identify anomalous human behaviour. Upon detection, alerts are automatically triggered and sent to administrators for timely intervention. The solution integrates scene classification, action recognition, and notification services into a unified platform. By automating surveillance analysis, the system enhances situational awareness and reduces manual monitoring efforts. This research aims to build a scalable, efficient, and adaptable security solution to improve public safety.
Video-based activity recognition is an evolving field of computer vision that leverages spatial and temporal features. While several methods have shown promising results in static pose estimation , applying deep networks to video sequences still presents challenges due to the added temporal dimension. To overcome this, our project exploits the strength of CNNs and motion-based frame analysis to capture and classify activity over time. Unlike earlier systems that rely on expensive hardware such as depth sensors with limitations like indoor-only use, our method is optimized for real-world environments with low-cost hardware and deployable on both desktop and mobile platforms.
Key Words: Suspicious activity detection, Deep
learning, Convolutional Neural Networks (CNN), Realtime surveillance, Computer vision, Crowd monitoring, Smart surveillance.
1.INTRODUCTION
The proposed system is structured into preprocessing, feature extraction, and classification modules, with optimized performance using Python (via Spyder IDE) on the Anaconda platform. Through automation and real-time processing, this model aims to reduce human effort while ensuring robust surveillance coverage. The output is user-friendly and adaptable, making it suitable for widespread deployment in both public and private security systems.
The rise in criminal activities and threats to public safety has led to increased deployment of video surveillance systems in sensitive areas such as malls, airports, railway stations, banks, and educational institutions. However, continuous manual monitoring of live video feeds is impractical and prone to fatigueinduced errors. To address this, we propose an intelligent surveillance system capable of detecting suspicious human activity in real-time using deep learning and neural network techniques.
2. PROBLEM STATEMENT
Suspicious activity detection refers to identifying human postures, gestures, and actions that deviate from normal behavioral patterns. Traditional approaches in computer vision have largely focused on static images and lacked temporal awareness, which is essential for accurate interpretation of human behavior
© 2024, IRJET
|
Impact Factor value: 8.315
Public safety in densely populated environments— such as transportation hubs, commercial complexes, and educational institutions—demands constant surveillance to prevent and respond to suspicious or abnormal human activities. Traditional surveillance systems depend heavily on manual monitoring of real-
|
ISO 9001:2008 Certified Journal
|
Page 581