Intelligent Real-Time Suspicious Activity Detection Using Convolutional Neural Networks

Abstract

Ensuring public safety has become an increasingly important challenge as urban
environments expand and security threats continue to rise. Traditional surveillance
systems rely heavily on human operators to continuously monitor video feeds, which is
inefficient, error-prone, and impractical for large-scale deployments. To address these
limitations, this project proposes an intelligent real-time suspicious activity detection
system based on Convolutional Neural Networks (CNNs). CNNs are highly effective in
automatically extracting and learning spatial features from images and video frames,
making them suitable for activity recognition tasks. The proposed system employs a
deep-learning-based architecture capable of identifying suspicious movements, gestures,
and behaviors without manual intervention.
The system processes incoming video frames from surveillance cameras, performs
preprocessing such as normalization and resizing, and then feeds them into a trained
CNN model. The CNN extracts high-level features and classifies activities into normal or
suspicious categories. By leveraging spatial feature learning and motion pattern
recognition, the system can detect activities such as fighting, theft attempts, aggressive
gestures, or unauthorized intrusions. The model is trained on publicly available human
activity datasets enriched with various classes of suspicious and non-suspicious
movements, ensuring a wide range of behavior recognition.