Enhancing Public Safety: Detection of Weapons and Violence in CCTV Videos with Deep Learning
Fiza Abdul Razzaq, Muhammad Abbas Chaudary, Sumaria Fareed, Warda Tariq, Muhammad Waqas, Shoaib Javaid
Abstract
Our research employs cutting-edge deep learning techniques to automate the detection of weapons and violent activities in CCTV footage. By utilizing advanced deep learning models, our system swiftly identifies violence, including fights, and detects weapons, enhancing public safety by generating real time alerts for relevant authorities. We have employed YOLOv5 for weapon detection and a combination of ResNet and Bi-LSTM for Violence Detection. After preprocessing and feature extraction, trained models can detect weapons and violent activities effectively. Evaluation on diverse datasets demonstrates strong performance. We have also demonstrated the effectiveness of proposed architecture on hockey fight dataset showing comparisons with state of the art models. We address real-world challenges like data biases and model generalization, emphasizing scalability through integration with law enforcement systems. In conclusion, our work contributes to automated detection with promising security applications.