GIS assisted Multi-Camera surveillance system with novel calibration approach for critical facilities

The protection of any strategic and critical facilities from physical intrusionsis of prime importance to the authorities. Now-a-days, a lot of video camerasare installed for surveillance to detect intrusions along the perimeter of suchfacilities. However, monitoring through video feed from a large number of camerassimultaneously by security personnel is tedious and often error prone. This paperdescribes an intelligent system developed to assist in surveillance by detectingintrusions from the video feed of already existing infrastructure of cameras andby marking the intrusion on a GIS map in real time. GIS mapping integratesthe output of multiple cameras in a single bird’s eye view to provide a pictureof entire perimeter in a glance. The intrusions are detected using deep learningbased object detection model and the pixel location of intrusions are transformedto GIS coordinates. This approach helps in reducing the number of watch towers,patrolling frequency etc., which in turn reduces the costs to security. This alsoacts as an assisting tool to the security by pointing at the right camera feed, inwhich a probable intrusion is occurring. This approach has been implementedat a critical facility with feed from geographically isolated multiple surveillancecameras located along the perimeter of 7km stretch. The video stream is collectedand processed at a centralised server with CUDA enabled GPU devices. Thesystem achieves a frame rate of 10fps for 75 cameras. This software architectureis modular and seamlessly scales-up to handle any number of cameras