Published Paper Details:

Surveillance systems play a crucial role in maintaining security and order in public places. Despite CCTV cameras installed at most of the locations, crimes that are captured in CCTVs not reported in an effective way. In this context, there has been an enhanced attention for increased surveillance in order to monitor crowd behaviour and to nab fugitives, suspects and criminals, especially in bus stations, railway stations and airports and other crowded locations. CCTV, in general, requires human supervision which may lead to missing some important crime events due to human error. The proposed project aims to help automate the process of monitoring the CCTV footage and supports in proactively detecting suspicious behaviour in real-time videos with alerts sent to higher authorities. The alert is generated and forwarded to designated authorities if suspicious events occur, which will help in preventing crimes, assisting in crime investigation, identification of suspects and also supports in automating some of the groundwork that is manual and time-consuming. Towards this, the paper attempts to incorporate two independently working machine learning models into one working model. To get the best results, transfer learning has been used for enhanced performance of each model and give better and faster results. The video feed from the CCTV will be taken as the input in our system and image frames will be extracted and fed into the two sub-systems of the project. For criminal/suspects face recognition, transfer learning on VGG16 model is used, for weapon detection, darknet framework is to be applied with YOLOv4 algorithm to get the desired object detection. These machine learning systems have been designed to work in parallel which in turn ensures faster output.