Crime Anomaly Detection using CNN and Ensemble Model

Abstract

Every single day, thousands of crimes are perpetrated, and hundreds may be probably taking place right now throughout the world. Without a doubt, crime is viewed as a social blight. Nothing can truly stop it, no matter what is done. Surveillance cameras, on the other hand, can dramatically minimize it. Using public surveillance camera systems to prevent, document, and minimize crime can be a cost-effective solution. Installing enough cameras to detect crimes in progress and integrating technology to automate the monitoring of the live stream from these cameras will result in the most effective systems. Because of its self-learning characteristics, the advanced Artificial Intelligence surveillance system is constantly learning and improving. The Deep Learning Algorithms applied in this work processes videos using electronic devices like cameras in real-time termed as image processing, saving both human resources and a great deal of time. The highest accuracy of 86.6% was attained by Ensemble Model, followed by Inception Model with SGD Optimizer, Leaky Relu Activation Function giving an accuracy of 83.43%. Hence, anomalies were detected efficiently using decision making in real-time surveillance scenarios.