Smart Surveillance for Automated Violence Detection in Public Spaces Using Deep Learning

Abstract - In contemporary security systems, automated surveillance has become essential for maintaining public safety. An enhanced real-time violence detection system that uses deep learning approaches is presented in this work. The suggested framework successfully detects violent activity in video streams by using Convolutional Neural Networks (CNN) with MobileNetV2 for the extraction of spatial features and Bidirectional Long Short-Term Memory (Bi-LSTM) for the study of temporal sequences. The system, which was developed with Python, TensorFlow, and OpenCV, incorporates an alert mechanism that sounds an alarm when it detects violence. According to experimental assessments, the system's excellent precision and efficiency make it a good fit for smart surveillance applications in both public and private settings. Key Words: OpenCV, CNN, LSTM, MobileNetV2, real-time surveillance, security systems, deep learning, and violence detection.