Crime Scene Prediction by Detecting Threatening Objects using Deep Learning Techniques

Abstract

Last couple of decades security has been a pressing issue and to overcome this, surveillance cameras are installed in several public places which helps in segregating crimes. Surveillance cameras allow us to watch and helps in creating a secure society. The data captured using cameras play a vital role in monitoring, predicting events and goal-driven analysis applications including anomalies and intrusion detection. The process followed in providing the input on the crime is to analyze the frames captured by the surveillance cameras and detecting the anomalies and sending an alert message to concerned authorities. This paper aims to put forth a unique method for anomaly detection based on deep learning techniques, which is designed by studying various existing models. Max pooling and ReLU are used in Convolutional Neural Network (CNN). Max pooling is a pooling operation that calculates the maximum, or largest, value in the patch of each feature map. This method has been evaluated using UCSD dataset and showed an increase in accuracy. Incorporating such techniques can help in crime detection at an early stage.