Anomaly Detection in Surveillance Videos Using Machine Learning

The primary objective of this study is to increase public safety by employing machine learning to identify security hazards including theft, violence, and accidents in CCTV footage. The system makes use of Long Short-Term Memory (LSTM) networks and convolutional neural networks (CNNs) that were trained using the University of Central Florida (UCF) Crime dataset. Sparsity and temporal smoothness restrictions are used in an anomaly-ranking framework to improve accuracy and reduce false positives. The system ensures real-time contact with security personnel by integrating Twilio for Short Message Service (SMS) alerts and Simple Mail Transfer Protocol (SMTP) with Multipurpose Internet Mail Extensions (MIME) for email notifications. Through transfer learning, data augmentation, and hyperparameter optimization, performance is further enhanced. This scalable, real-time system offers a productive method for detecting anomalies in crowded, high-risk settings.