Crime is an increasing threat to our society and poses a significant threat to public safety and social well-being. With the advent of faster and improved technology we should be able to leverage this to our advantage, it should be used to help police detect crimes and ensure public safety. Traditionally, crime prevention and response mechanisms have relied heavily on human monitoring of surveillance systems, which is susceptible to limitations such as oversight, fatigue, and delayed reaction times. In addition to this there is a lot of human resources wasted if everytime there is a call for help and police have to respond as they have no way of verifying if the call is indeed an emergency or not. To counter these challenges I intend to propose a Crime Surveillance System (CSS) which aims to automate the detection of criminal activities, overcoming the limitations of human monitoring and enabling proactive response measures. Even a momentary lapse could be the difference between stopping a crime as it's happening versus being too late. I will be using 2 main Deep Learning algorithms for this project and those are: MobileNetV2 and ResNet50V2, combining these 2 models to make an ensemble model that will have better accuracy and will help the law enforcement officers in many ways.