Violence Detection in Surveillance Systems Using MobileNetV2 and Deep Learning
Dhiraj Wagh, Janhavi Hire, Mayuri Phad, Ajay Bhosale, Riddhi Mirajkar, Anuradha Yenkikar · 2024
This paper addresses the critical need for efficient real time violence detection in security systems. Traditional methods often lack the ability to respond promptly and accurately in dynamic environments. To overcome this challenge, we developed a deep learning-based system integrated with cloud technology and instant alert mechanisms. The system utilizes OpenCV for video processing and the MTCNN framework for face detection, leveraging a pre-trained MobileNetv2 model to identify violent content in live video streams. Our model achieved an accuracy of 95.76% with a precision of 94.25%, recall of 93.50% and F1-score of 93.87% in detecting Violent incident upon detection, the system extract key frames, enhances them for clarity, and uploads them to Firebase Storage. Simultaneously, the system gathers the device's geolocation and sends a notification with the image and location to specified Telegram group via bot. Extensive testing demonstrates the system's effectiveness in detecting Violent incidents, providing a reliable tool for enhancing public safety. Future enhancements could focus on expanding its deployment to larger networks, improving detection accuracy.