Knife Detection using Indoor Surveillance Camera

Abstract

Security and safeness are constant topics to be solved in cities around the world, finding ways to detect weapons threatening human beings is an important challenge. This paper presents an approach for solving knife detection in Close Circuit Television (CCTV) cameras videos. In this sense, knife detection is the goal in this work and through a combination of color and invariant moments techniques, the system presented reaches the objective and turns reliable in indoor environments, with or without environmental illumination. The work has been proved under white and infrared lighting, with a range between 0.5 to 4.0 meters of distance from camera to knives with positive qualities of detection. Supporting this work, videos were uploaded on web. The reported error in them runs from 0% to 1.192%. This system may be useful at convenience stores, banks, theatres and some others public places with commercial surveillance cameras from a relatively long distance. Results offer a simple classification because to important features found. For instance, when the system may be executed by parallel processors or in pipeline method it could be detecting more than one knife on scene.