Firearm Detection from Surveillance Cameras Using Image Processing and Machine Learning Techniques

Abstract

The increasing number of terrorist acts and lone wolf attacks on places of public gathering such as Hotels and Cinemas has solidified the need for much denser Closed-circuit Television (CCTV) systems. The increasing number of CCTV cameras has deemed it almost impossible for a human operator to inspect all the video streams and detect possible terror events. One of the common types of terror event is called “Active Shooter”. Events such as the 2008 Mumbai shooting, shooting at the movie theater in Colorado (USA), Oslo (Norway) and recently an attacker opened gun fire at an outdoor music festival in Las Vegas on Oct. 1, 2017, USA. Therefore in this work, the detection of an “Active Shooter” carrying a non-concealed firearm and alerting the CCTV operator of a potentially dangerous event both visually and audibly has been carried out. The proposed approach of gun detection uses a feature extraction techniques and a convolutional neural network classifier for classifying objects as either a gun or not a gun. And the classification accuracy achieved by the proposed approach is 97.78%.