Anomaly Detection Using Convolutional Neural Networks for Crime Prevention: A Deep Learning Approach
Ms. Noor Unnisa, Mohammed Ehtesham Ul baqui, Ms. Vibhavari N, Meherin Sultana, Shaik Irfan, Mohammed Ehtesham Ul baqui · International Journal of Latest Technology in Engineering Management & Applied Science · 2025
Abstract: Anomaly detection using Convolutional Neural Networks (CNNs) has emerged as a powerful tool for identifying criminal activities, including robberies, assaults, and homicides, within surveillance environments. This research presents a deep learning-based framework that leverages CNNs for spatial feature extraction and combines them with temporal modelling to recognize irregular behaviour in public safety contexts. By analysing surveillance footage and sensor-based data, the system detects anomalies in movement patterns, crowd density, and object interactions, thereby aiding in real-time threat assessment and crime prevention. The proposed method utilizes pre-trained CNN models for high-level visual representation and integrates hybrid approaches, such as CNN-LSTM and 3D CNNs, to capture spatiotemporal dynamics of suspicious activities. Our framework is tested on benchmark datasets like UCF-Crime and UAV surveillance feeds, achieving high accuracy in detecting abnormal behaviour. Furthermore, anomaly detection is enhanced using advanced feature extraction techniques and real-time classification mechanisms tailored for smart city surveillance systems. This study contributes to the evolving field of AI-driven public safety by integrating CNN architectures with context-aware, temporal anomaly detection techniques, enabling proactive criminal activity prediction and continuous learning from dynamic environments. The results demonstrate the potential of deep neural networks to support law enforcement agencies in real-time monitoring, ultimately reducing crime response times and enhancing urban security infrastructures.