Suspicious Activity Detection Using Deep Learning Approach

Abstract

This project is focused on improving security systems by using deep learning and computer vision to spot suspicious human actions in real time. The goal is to help detect unsafe or dangerous behaviour by using smart models that can understand human movements from video footage. One of the main models used in this system is called Slow-Fast. This model looks at both slow and fast video frames to better understand how people move over time. By doing this, it can catch even small or complex actions that might seem unusual or suspicious. To make activity recognition more accurate, the system also uses another model called ResNet50, which helps in telling the difference between normal and strange behaviors by learning from patterns in the video. For example, YOLOv5 is used to quickly and accurately detect dangerous items like weapons or events like accidents and explosions. This helps in sending quick alerts during emergencies. Another part of the system uses MediaPipe, a tool that studies human body movements. It tracks how people move and can spot physical fights by noticing aggressive or violent actions. By putting all these tools together, the system can watch over places like schools, public areas, or offices and quickly alert security teams if something risky is happening. The project shows how advanced technology can be used in smart and helpful ways to keep people safe.