Real-Time Violence Detection and Alert System

Abstract —Personal safety and quality of life are seriously threatened by the violent acts. In order to curb the hostile conduct, numerous other strategies have been implemented, such as installing and maintaining surveillance systems. It will be extremely relevant if the monitoring systems are able to recognize violent conduct on their own and immediately send warning or alarm signals to the authorities. New researchers are drawn to the active study fields of computer vision and image processing for the detection of aggressive and aberrant conduct. In order to detect violence in video streams, this research suggests a real-time violence detection and alarm system that uses MobileNetV2 along with LSTM. The system offers a scalable, precise, and flexible public safety solution by utilizing a Telegram Bot for real-time alerting message. Keywords — MobileNetV2, Realtime detection, LSTM, Transfer learning, Neural network, callbacks, RLVS.