Vision-Based Crime Detection System

Abstract

Nowadays, crime is one of the major issues in Pakistan. Banks/houses have installed CCTV cameras for recordings, but these cameras cannot detect any crime in real-time and cannot resist the crime by informing to Police station. CCTV cameras can also be used to detect crimes. According to the “CCTV Operational Needed Things Manual 2009”, the responsibility of the CCTV controller is to observe, identify, control, and detect situations and people that may be unsafe to other people or their property. The task is very difficult because the footage contains a lot of information with major differences and challenges. All security systems still require human supervision, and it is essential to get the attention of CCTV officers to analyse the situation. However, many detection systems have been proposed, such as the Robbery/Theft Detection System, Automatic Handgun Detection System, and Firearms and Knives Detection System using CCTV cameras. A vision-based crime detection system is being proposed to detect criminal behaviour. One of the best existing models used for the object detection algorithm would be decided on runtime and data collection for creating the dataset; then, the dataset will be annotated for training the object detection algorithm. The system will use a CCTV camera to take the stream and will detect crime behaviours if occurred. The system has an admin portal (house owner/bank manager) that can observe, identify, control, and detect situations of people that may be unsafe to other people or their property. Moreover, if any crime occurs, an alert message will be sent to the nearby police station and owner with the address and detected image URL.