Crime Detection and Geospatial Alert System using LRCN and Streamlit for Smart City Surveillance

Advances in surveillance cameras that have risen with the need for increased security and prevention of real- time crimes have become very common in public. Advanced analyses in machine learning can detect crime from video footage, but problems persist with the current systems. Among the problems include being unable to pinpoint specific criminal activities and a lack of location-based alerts or real-time analysis. Most of the systems that exist cannot perform video feed analysis in real-time video feeds or cannot send alerts to relevant authorities promptly. As a result, such a solution would be of limited effectiveness for public safety. All these new techniques, such as CNN and RNN, have been implemented on crimes regarding action recognition and video surveillance. However, these approaches are often limited in precision, detection speed, and scalability. Specifically, a possible identified limitation is the precise detection of criminal activities within varied environments with a real-time response. Moreover, geocoding for localized alerts has not yet been well applied within the current systems. This paper presents a real-time crime detection system that uses the Long-term Recurrent Convolutional Network model to analyze live feeds from video surveillance systems. The system is meant to classify actions into criminal and noncriminal actions with a very good success rate of 98%. Even as criminal activity is detected, the system provides real-time alerts while storing video clips as evidence. Improving public safety and offering proactive crime prevention is a significant improvement that results from tackling issues of accuracy, real-time processing, and specificity of alerts by this system. Real-time analysis, generation of alerts, and evidence storage integrate into a solid solution in enhancing the security of public space.