Crime detection and prevention are critical aspects of maintaining public safety and security. Law enforcement agencies are often faced with the challenge of identifying and responding to criminal activities in a timely manner. Traditional methods of crime analysis rely on historical data and patterns to predict and prevent crime which are highly redundant and not very efficient. The objective is to design and implement an advanced real-time monitoring system capable of efficiently and accurately identifying various abnormal events—crimes. The anomaly detection entails identifying abnormal events that deviate from regular patterns within sequences of video. The developed real-time anomaly (crime) detection system which involves crimes like: Abuse, Arrest, Arson, Violence, Burglary, Explosion, Fighting, Robbery, Shooting, Stealing, Shoplifting, and Vandalism which are specifically chosen for public safety. Upon detecting anomalous activities, alarm is raised to warn local authorities and alert the public. To detect anomalies in videos, the system employs I3D-ResNet 50 for feature extraction and MGFN for anomaly detection and YOLOv3 for object detection from the tagged anomalous video list. The proposed MGFN system is validated using UCF-Crime dataset and compared with other techniques including standard supervised, unsupervised, and weakly-supervised algorithms. Constantly analyzing security camera footage with an accuracy of 85% higher than current standard methods allows for continuous monitoring and early detection of suspicious behaviors.