AI-Driven Intelligent Surveillance System Using YOLOv8, DeepSORT, and Vision Transformers

This project develops an AI-based surveillance system that can automatically detect people,
track their movements, and identify security risks in real time. The research aims to combine
object detection (YOLOv8), tracking (DeepSORT), and transformer-based scene understanding
(ViT) to test the hypothesis that explainable attention maps can improve the accuracy and
clarity of surveillance decisions. The main experiments include detecting individuals in video
frames, generating attention heatmaps, analyzing loitering or crowding patterns, and
assigning risk levels based on density and model confidence. If successful, the system will
provide a reliable and interpretable surveillance tool with strong applications in public safety,
campus monitoring, and smart security environments.