Computer Vision Based Campus Surveillance

Security is an important aspect for all of the human race and the organizations that are run by them. In this paper, an automated multipurpose security and surveillance system is proposed which is useful at highly critical places such as borders or highly restricted areas where tracking of each and every object is important. This system works on real time video footage captured using drone cameras or Closed Circuit Television (CCTV) systems and with the use deep learning object detection techniques detects buildings, trees, vehicles, water bodies, playground and slums. The system uses Faster R-CNN architecture and with the help of transfer learning the top layers of the architecture are fine tuned to detect system specific objects at a high accuracy. The video frames are given as input to convolution neural network (CNN) layers for classification of live footage that reveals the count of all objects detected. It approximately gives the percentage of each object class identified.