Suspicious Activity Detection based on Audio Detecting Methodology using Deep Learning

Abstract

Suspicious Activity refer to actions that appear unusual or questionable and may indicate the possibility of potentially illegal, harmful or illicit activities. These activities can be an early warning sign of criminal activity and their detection and prevention is necessary. This in turn helps in protection of assets, protection of individuals and prevention of the crime. The already increased rate of crime causes a lot of significant economic and personal damage. For detection, the counter is often standard human surveillance at all times. A smart surveillance system should be able to identify these activities through any means. In this paper, we propose a method using Deep Learning with Tensorflow to classify suspicious sounds like gun shots, jackhammer or glass breaking. A sophisticated microphone can be placed in Areas of interest like a Bank Locker or ATM and using the model, incoming sounds can be detected and classified. The Deep Learning Models that are trained, evaluated and tested are a Dense Model and an LSTM Model.The LSTM Model performs the best for identification of the sounds with an accuracy of 96.02%.